Mercurial > pylearn
annotate dataset.py @ 98:7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
author | Frederic Bastien <bastienf@iro.umontreal.ca> |
---|---|
date | Tue, 06 May 2008 13:50:54 -0400 |
parents | 6fe972a7393c |
children | a8da709eb6a9 |
rev | line source |
---|---|
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
1 |
12
ff4e551490f1
Added LookupList type in lookup_list.py and used it to keep order
bengioy@esprit.iro.umontreal.ca
parents:
11
diff
changeset
|
2 from lookup_list import LookupList |
ff4e551490f1
Added LookupList type in lookup_list.py and used it to keep order
bengioy@esprit.iro.umontreal.ca
parents:
11
diff
changeset
|
3 Example = LookupList |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
4 from misc import unique_elements_list_intersection |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
5 from string import join |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
6 from sys import maxint |
45
a5c70dc42972
Test functions for dataset.py
bengioy@grenat.iro.umontreal.ca
parents:
44
diff
changeset
|
7 import numpy |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
8 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
9 class AbstractFunction (Exception): """Derived class must override this function""" |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
10 class NotImplementedYet (NotImplementedError): """Work in progress, this should eventually be implemented""" |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
11 |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
12 class DataSet(object): |
16 | 13 """A virtual base class for datasets. |
14 | |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
15 A DataSet can be seen as a generalization of a matrix, meant to be used in conjunction |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
16 with learning algorithms (for training and testing them): rows/records are called examples, and |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
17 columns/attributes are called fields. The field value for a particular example can be an arbitrary |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
18 python object, which depends on the particular dataset. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
19 |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
20 We call a DataSet a 'stream' when its length is unbounded (otherwise its __len__ method |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
21 should return sys.maxint). |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
22 |
16 | 23 A DataSet is a generator of iterators; these iterators can run through the |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
24 examples or the fields in a variety of ways. A DataSet need not necessarily have a finite |
16 | 25 or known length, so this class can be used to interface to a 'stream' which |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
26 feeds on-line learning (however, as noted below, some operations are not |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
27 feasible or not recommanded on streams). |
16 | 28 |
29 To iterate over examples, there are several possibilities: | |
90
a289b8bed64c
corrected comment
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
88
diff
changeset
|
30 - for example in dataset: |
a289b8bed64c
corrected comment
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
88
diff
changeset
|
31 - for val1,val2,... in dataset: |
a289b8bed64c
corrected comment
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
88
diff
changeset
|
32 - for example in dataset(field1, field2,field3, ...): |
a289b8bed64c
corrected comment
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
88
diff
changeset
|
33 - for val1,val2,val3 in dataset(field1, field2,field3): |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
34 - for minibatch in dataset.minibatches([field1, field2, ...],minibatch_size=N): |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
35 - for mini1,mini2,mini3 in dataset.minibatches([field1, field2, field3], minibatch_size=N): |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
36 Each of these is documented below. All of these iterators are expected |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
37 to provide, in addition to the usual 'next()' method, a 'next_index()' method |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
38 which returns a non-negative integer pointing to the position of the next |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
39 example that will be returned by 'next()' (or of the first example in the |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
40 next minibatch returned). This is important because these iterators |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
41 can wrap around the dataset in order to do multiple passes through it, |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
42 in possibly unregular ways if the minibatch size is not a divisor of the |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
43 dataset length. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
44 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
45 To iterate over fields, one can do |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
46 - for field in dataset.fields(): |
46
c5b07e87b0cb
comments modif made by Yoshua
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
45
diff
changeset
|
47 for field_value in field: # iterate over the values associated to that field for all the dataset examples |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
48 - for field in dataset(field1,field2,...).fields() to select a subset of fields |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
49 - for field in dataset.fields(field1,field2,...) to select a subset of fields |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
50 and each of these fields is iterable over the examples: |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
51 - for field_examples in dataset.fields(): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
52 for example_value in field_examples: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
53 ... |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
54 but when the dataset is a stream (unbounded length), it is not recommanded to do |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
55 such things because the underlying dataset may refuse to access the different fields in |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
56 an unsynchronized ways. Hence the fields() method is illegal for streams, by default. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
57 The result of fields() is a DataSetFields object, which iterates over fields, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
58 and whose elements are iterable over examples. A DataSetFields object can |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
59 be turned back into a DataSet with its examples() method:: |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
60 dataset2 = dataset1.fields().examples() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
61 and dataset2 should behave exactly like dataset1 (in fact by default dataset2==dataset1). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
62 |
16 | 63 Note: Fields are not mutually exclusive, i.e. two fields can overlap in their actual content. |
64 | |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
65 Note: The content of a field can be of any type. Field values can also be 'missing' |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
66 (e.g. to handle semi-supervised learning), and in the case of numeric (numpy array) |
46
c5b07e87b0cb
comments modif made by Yoshua
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
45
diff
changeset
|
67 fields (i.e. an ArrayFieldsDataSet), NaN plays the role of a missing value. |
c5b07e87b0cb
comments modif made by Yoshua
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
45
diff
changeset
|
68 What about non-numeric values? None. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
69 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
70 Dataset elements can be indexed and sub-datasets (with a subset |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
71 of examples) can be extracted. These operations are not supported |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
72 by default in the case of streams. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
73 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
74 - dataset[:n] returns a dataset with the n first examples. |
16 | 75 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
76 - dataset[i1:i2:s] returns a dataset with the examples i1,i1+s,...i2-s. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
77 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
78 - dataset[i] returns an Example. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
79 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
80 - dataset[[i1,i2,...in]] returns a dataset with examples i1,i2,...in. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
81 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
82 - dataset[fieldname] an iterable over the values of the field fieldname across |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
83 the dataset (the iterable is obtained by default by calling valuesVStack |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
84 over the values for individual examples). |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
85 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
86 - dataset.<property> returns the value of a property associated with |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
87 the name <property>. The following properties should be supported: |
41 | 88 - 'description': a textual description or name for the dataset |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
89 - 'fieldtypes': a list of types (one per field) |
78 | 90 A DataSet may have other attributes that it makes visible to other objects. These are |
91 used to store information that is not example-wise but global to the dataset. | |
92 The list of names of these attributes is given by the attribute_names() method. | |
41 | 93 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
94 Datasets can be concatenated either vertically (increasing the length) or |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
95 horizontally (augmenting the set of fields), if they are compatible, using |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
96 the following operations (with the same basic semantics as numpy.hstack |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
97 and numpy.vstack): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
98 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
99 - dataset1 | dataset2 | dataset3 == dataset.hstack([dataset1,dataset2,dataset3]) |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
100 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
101 creates a new dataset whose list of fields is the concatenation of the list of |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
102 fields of the argument datasets. This only works if they all have the same length. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
103 |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
104 - dataset1 & dataset2 & dataset3 == dataset.vstack([dataset1,dataset2,dataset3]) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
105 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
106 creates a new dataset that concatenates the examples from the argument datasets |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
107 (and whose length is the sum of the length of the argument datasets). This only |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
108 works if they all have the same fields. |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
109 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
110 According to the same logic, and viewing a DataSetFields object associated to |
46
c5b07e87b0cb
comments modif made by Yoshua
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
45
diff
changeset
|
111 a DataSet as a kind of transpose of it, fields1 & fields2 concatenates fields of |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
112 a DataSetFields fields1 and fields2, and fields1 | fields2 concatenates their |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
113 examples. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
114 |
41 | 115 A dataset can hold arbitrary key-value pairs that may be used to access meta-data |
116 or other properties of the dataset or associated with the dataset or the result | |
117 of a computation stored in a dataset. These can be accessed through the [key] syntax | |
118 when key is a string (or more specifically, neither an integer, a slice, nor a list). | |
78 | 119 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
120 A DataSet sub-class should always redefine the following methods: |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
121 - __len__ if it is not a stream |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
122 - fieldNames |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
123 - minibatches_nowrap (called by DataSet.minibatches()) |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
124 - valuesHStack |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
125 - valuesVStack |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
126 For efficiency of implementation, a sub-class might also want to redefine |
72
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
127 - hasFields |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
128 - __getitem__ may not be feasible with some streams |
2b6656b2ef52
Changed docs slightly
Joseph Turian <turian@iro.umontreal.ca>
parents:
66
diff
changeset
|
129 - __iter__ |
78 | 130 A sub-class should also append attributes to self._attribute_names |
131 (the default value returned by attributeNames()). | |
132 By convention, attributes not in attributeNames() should have a name | |
133 starting with an underscore. | |
134 @todo enforce/test that convention! | |
2
3fddb1c8f955
Rewrote DataSet interface and created FiniteDataSet interface.
bengioy@bengiomac.local
parents:
1
diff
changeset
|
135 """ |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
136 |
83 | 137 numpy_vstack = lambda fieldname,values: numpy.vstack(values) |
138 numpy_hstack = lambda fieldnames,values: numpy.hstack(values) | |
77
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
139 |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
140 def __init__(self,description=None,fieldtypes=None): |
41 | 141 if description is None: |
142 # by default return "<DataSetType>(<SuperClass1>,<SuperClass2>,...)" | |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
143 description = type(self).__name__ + " ( " + join([x.__name__ for x in type(self).__bases__]) + " )" |
41 | 144 self.description=description |
60 | 145 self.fieldtypes=fieldtypes |
78 | 146 self._attribute_names = ["description"] |
147 if fieldtypes: | |
148 self._attribute_names.append("fieldtypes") | |
149 | |
150 def attributeNames(self): return self._attribute_names | |
151 | |
152 def setAttributes(self,attribute_names,attribute_values): | |
153 for name,value in zip(attribute_names,attribute_values): | |
154 self.__setattr__(name,value) | |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
155 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
156 class MinibatchToSingleExampleIterator(object): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
157 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
158 Converts the result of minibatch iterator with minibatch_size==1 into |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
159 single-example values in the result. Therefore the result of |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
160 iterating on the dataset itself gives a sequence of single examples |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
161 (whereas the result of iterating over minibatches gives in each |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
162 Example field an iterable object over the individual examples in |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
163 the minibatch). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
164 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
165 def __init__(self, minibatch_iterator): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
166 self.minibatch_iterator = minibatch_iterator |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
167 self.minibatch = None |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
168 def __iter__(self): #makes for loop work |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
169 return self |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
170 def next(self): |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
171 size1_minibatch = self.minibatch_iterator.next() |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
172 if not self.minibatch: |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
173 self.minibatch = Example(size1_minibatch.keys(),[value[0] for value in size1_minibatch.values()]) |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
174 else: |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
175 self.minibatch._values = [value[0] for value in size1_minibatch.values()] |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
176 return self.minibatch |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
177 |
23
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
178 def next_index(self): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
179 return self.minibatch_iterator.next_index() |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
180 |
3
378b68d5c4ad
Added first (untested) version of ArrayDataSet
bengioy@bengiomac.local
parents:
2
diff
changeset
|
181 def __iter__(self): |
16 | 182 """Supports the syntax "for i in dataset: ..." |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
183 |
16 | 184 Using this syntax, "i" will be an Example instance (or equivalent) with |
185 all the fields of DataSet self. Every field of "i" will give access to | |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
186 a field of a single example. Fields should be accessible via |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
187 i["fielname"] or i[3] (in the order defined by the elements of the |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
188 Example returned by this iterator), but the derived class is free |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
189 to accept any type of identifier, and add extra functionality to the iterator. |
16 | 190 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
191 The default implementation calls the minibatches iterator and extracts the first example of each field. |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
192 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
193 return DataSet.MinibatchToSingleExampleIterator(self.minibatches(None, minibatch_size = 1)) |
2
3fddb1c8f955
Rewrote DataSet interface and created FiniteDataSet interface.
bengioy@bengiomac.local
parents:
1
diff
changeset
|
194 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
195 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
196 class MinibatchWrapAroundIterator(object): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
197 """ |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
198 An iterator for minibatches that handles the case where we need to wrap around the |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
199 dataset because n_batches*minibatch_size > len(dataset). It is constructed from |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
200 a dataset that provides a minibatch iterator that does not need to handle that problem. |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
201 This class is a utility for dataset subclass writers, so that they do not have to handle |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
202 this issue multiple times, nor check that fieldnames are valid, nor handle the |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
203 empty fieldnames (meaning 'use all the fields'). |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
204 """ |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
205 def __init__(self,dataset,fieldnames,minibatch_size,n_batches,offset): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
206 self.dataset=dataset |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
207 self.fieldnames=fieldnames |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
208 self.minibatch_size=minibatch_size |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
209 self.n_batches=n_batches |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
210 self.n_batches_done=0 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
211 self.next_row=offset |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
212 self.offset=offset |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
213 self.L=len(dataset) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
214 assert offset+minibatch_size<=self.L |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
215 ds_nbatches = (self.L-self.next_row)/self.minibatch_size |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
216 if n_batches is not None: |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
217 ds_nbatches = min(n_batches,ds_nbatches) |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
218 if fieldnames: |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
219 assert dataset.hasFields(*fieldnames) |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
220 else: |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
221 self.fieldnames=dataset.fieldNames() |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
222 self.iterator = self.dataset.minibatches_nowrap(self.fieldnames,self.minibatch_size, |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
223 ds_nbatches,self.next_row) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
224 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
225 def __iter__(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
226 return self |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
227 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
228 def next_index(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
229 return self.next_row |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
230 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
231 def next(self): |
43
e92244f30116
Corrected iterator logic errors
bengioy@grenat.iro.umontreal.ca
parents:
42
diff
changeset
|
232 if self.n_batches and self.n_batches_done==self.n_batches: |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
233 raise StopIteration |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
234 upper = self.next_row+self.minibatch_size |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
235 if upper <=self.L: |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
236 minibatch = self.iterator.next() |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
237 else: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
238 if not self.n_batches: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
239 raise StopIteration |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
240 # we must concatenate (vstack) the bottom and top parts of our minibatch |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
241 # first get the beginning of our minibatch (top of dataset) |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
242 first_part = self.dataset.minibatches_nowrap(self.fieldnames,self.L-self.next_row,1,self.next_row).next() |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
243 second_part = self.dataset.minibatches_nowrap(self.fieldnames,upper-self.L,1,0).next() |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
244 minibatch = Example(self.fieldnames, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
245 [self.dataset.valuesVStack(name,[first_part[name],second_part[name]]) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
246 for name in self.fieldnames]) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
247 self.next_row=upper |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
248 self.n_batches_done+=1 |
43
e92244f30116
Corrected iterator logic errors
bengioy@grenat.iro.umontreal.ca
parents:
42
diff
changeset
|
249 if upper >= self.L and self.n_batches: |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
250 self.next_row -= self.L |
98
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
251 ds_nbatches = (self.L-self.next_row)/self.minibatch_size |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
252 if self.n_batches is not None: |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
253 ds_nbatches = min(self.n_batches,ds_nbatches) |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
254 self.iterator = self.dataset.minibatches_nowrap(self.fieldnames,self.minibatch_size, |
7186e4f502d1
bugfix in DataSet.minibatch to correctly wrap in all corner case.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
95
diff
changeset
|
255 ds_nbatches,self.next_row) |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
256 return DataSetFields(MinibatchDataSet(minibatch,self.dataset.valuesVStack, |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
257 self.dataset.valuesHStack), |
74
b4159cbdc06b
Fixed errors raised by test_dataset
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
73
diff
changeset
|
258 minibatch.keys()) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
259 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
260 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
261 minibatches_fieldnames = None |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
262 minibatches_minibatch_size = 1 |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
263 minibatches_n_batches = None |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
264 def minibatches(self, |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
265 fieldnames = minibatches_fieldnames, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
266 minibatch_size = minibatches_minibatch_size, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
267 n_batches = minibatches_n_batches, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
268 offset = 0): |
6
d5738b79089a
Removed MinibatchIterator and instead made minibatch_size a field of all DataSets,
bengioy@bengiomac.local
parents:
5
diff
changeset
|
269 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
270 Return an iterator that supports three forms of syntax: |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
271 |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
272 for i in dataset.minibatches(None,**kwargs): ... |
16 | 273 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
274 for i in dataset.minibatches([f1, f2, f3],**kwargs): ... |
16 | 275 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
276 for i1, i2, i3 in dataset.minibatches([f1, f2, f3],**kwargs): ... |
16 | 277 |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
278 Using the first two syntaxes, "i" will be an indexable object, such as a list, |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
279 tuple, or Example instance. In both cases, i[k] is a list-like container |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
280 of a batch of current examples. In the second case, i[0] is |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
281 list-like container of the f1 field of a batch current examples, i[1] is |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
282 a list-like container of the f2 field, etc. |
2
3fddb1c8f955
Rewrote DataSet interface and created FiniteDataSet interface.
bengioy@bengiomac.local
parents:
1
diff
changeset
|
283 |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
284 Using the first syntax, all the fields will be returned in "i". |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
285 Using the third syntax, i1, i2, i3 will be list-like containers of the |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
286 f1, f2, and f3 fields of a batch of examples on each loop iteration. |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
287 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
288 The minibatches iterator is expected to return upon each call to next() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
289 a DataSetFields object, which is a LookupList (indexed by the field names) whose |
80 | 290 elements are iterable and indexable over the minibatch examples, and which keeps a pointer to |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
291 a sub-dataset that can be used to iterate over the individual examples |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
292 in the minibatch. Hence a minibatch can be converted back to a regular |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
293 dataset or its fields can be looked at individually (and possibly iterated over). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
294 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
295 PARAMETERS |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
296 - fieldnames (list of any type, default None): |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
297 The loop variables i1, i2, i3 (in the example above) should contain the |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
298 f1, f2, and f3 fields of the current batch of examples. If None, the |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
299 derived class can choose a default, e.g. all fields. |
16 | 300 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
301 - minibatch_size (integer, default 1) |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
302 On every iteration, the variables i1, i2, i3 will have |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
303 exactly minibatch_size elements. e.g. len(i1) == minibatch_size |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
304 |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
305 - n_batches (integer, default None) |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
306 The iterator will loop exactly this many times, and then stop. If None, |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
307 the derived class can choose a default. If (-1), then the returned |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
308 iterator should support looping indefinitely. |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
309 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
310 - offset (integer, default 0) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
311 The iterator will start at example 'offset' in the dataset, rather than the default. |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
312 |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
313 Note: A list-like container is something like a tuple, list, numpy.ndarray or |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
314 any other object that supports integer indexing and slicing. |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
315 |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
316 """ |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
317 return DataSet.MinibatchWrapAroundIterator(self,fieldnames,minibatch_size,n_batches,offset) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
318 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
319 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
320 """ |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
321 This is the minibatches iterator generator that sub-classes must define. |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
322 It does not need to worry about wrapping around multiple times across the dataset, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
323 as this is handled by MinibatchWrapAroundIterator when DataSet.minibatches() is called. |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
324 The next() method of the returned iterator does not even need to worry about |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
325 the termination condition (as StopIteration will be raised by DataSet.minibatches |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
326 before an improper call to minibatches_nowrap's next() is made). |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
327 That next() method can assert that its next row will always be within [0,len(dataset)). |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
328 The iterator returned by minibatches_nowrap does not need to implement |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
329 a next_index() method either, as this will be provided by MinibatchWrapAroundIterator. |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
330 """ |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
331 raise AbstractFunction() |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
332 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
333 def __len__(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
334 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
335 len(dataset) returns the number of examples in the dataset. |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
336 By default, a DataSet is a 'stream', i.e. it has an unbounded length (sys.maxint). |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
337 Sub-classes which implement finite-length datasets should redefine this method. |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
338 Some methods only make sense for finite-length datasets. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
339 """ |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
340 return sys.maxint |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
341 |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
342 def is_unbounded(self): |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
343 """ |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
344 Tests whether a dataset is unbounded (e.g. a stream). |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
345 """ |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
346 return len(self)==sys.maxint |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
347 |
26
672fe4b23032
Fixed dataset errors so that _test_dataset.py works again.
bengioy@grenat.iro.umontreal.ca
parents:
23
diff
changeset
|
348 def hasFields(self,*fieldnames): |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
349 """ |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
350 Return true if the given field name (or field names, if multiple arguments are |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
351 given) is recognized by the DataSet (i.e. can be used as a field name in one |
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
352 of the iterators). |
29
46c5c90019c2
Changed apply_function so that it propagates methods of the source.
bengioy@grenat.iro.umontreal.ca
parents:
28
diff
changeset
|
353 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
354 The default implementation may be inefficient (O(# fields in dataset)), as it calls the fieldNames() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
355 method. Many datasets may store their field names in a dictionary, which would allow more efficiency. |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
356 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
357 return len(unique_elements_list_intersection(fieldnames,self.fieldNames()))>0 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
358 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
359 def fieldNames(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
360 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
361 Return the list of field names that are supported by the iterators, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
362 and for which hasFields(fieldname) would return True. |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
363 """ |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
364 raise AbstractFunction() |
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
365 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
366 def __call__(self,*fieldnames): |
23
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
367 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
368 Return a dataset that sees only the fields whose name are specified. |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
369 """ |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
370 assert self.hasFields(*fieldnames) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
371 return self.fields(*fieldnames).examples() |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
372 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
373 def fields(self,*fieldnames): |
29
46c5c90019c2
Changed apply_function so that it propagates methods of the source.
bengioy@grenat.iro.umontreal.ca
parents:
28
diff
changeset
|
374 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
375 Return a DataSetFields object associated with this dataset. |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
376 """ |
74
b4159cbdc06b
Fixed errors raised by test_dataset
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
73
diff
changeset
|
377 return DataSetFields(self,fieldnames) |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
378 |
2
3fddb1c8f955
Rewrote DataSet interface and created FiniteDataSet interface.
bengioy@bengiomac.local
parents:
1
diff
changeset
|
379 def __getitem__(self,i): |
28
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
380 """ |
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
381 dataset[i] returns the (i+1)-th example of the dataset. |
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
382 dataset[i:j] returns the subdataset with examples i,i+1,...,j-1. |
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
383 dataset[i:j:s] returns the subdataset with examples i,i+2,i+4...,j-2. |
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
384 dataset[[i1,i2,..,in]] returns the subdataset with examples i1,i2,...,in. |
41 | 385 dataset['key'] returns a property associated with the given 'key' string. |
386 If 'key' is a fieldname, then the VStacked field values (iterable over | |
387 field values) for that field is returned. Other keys may be supported | |
388 by different dataset subclasses. The following key names are encouraged: | |
389 - 'description': a textual description or name for the dataset | |
390 - '<fieldname>.type': a type name or value for a given <fieldname> | |
1
2cd82666b9a7
Added statscollector and started writing dataset and learner.
bengioy@esprit.iro.umontreal.ca
parents:
0
diff
changeset
|
391 |
39 | 392 Note that some stream datasets may be unable to implement random access, i.e. |
393 arbitrary slicing/indexing | |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
394 because they can only iterate through examples one or a minibatch at a time |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
395 and do not actually store or keep past (or future) examples. |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
396 |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
397 The default implementation of getitem uses the minibatches iterator |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
398 to obtain one example, one slice, or a list of examples. It may not |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
399 always be the most efficient way to obtain the result, especially if |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
400 the data are actually stored in a memory array. |
28
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
401 """ |
41 | 402 # check for an index |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
403 if type(i) is int: |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
404 return DataSet.MinibatchToSingleExampleIterator( |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
405 self.minibatches(minibatch_size=1,n_batches=1,offset=i)).next() |
41 | 406 rows=None |
407 # or a slice | |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
408 if type(i) is slice: |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
409 if not i.start: i.start=0 |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
410 if not i.step: i.step=1 |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
411 if i.step is 1: |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
412 return self.minibatches(minibatch_size=i.stop-i.start,n_batches=1,offset=i.start).next().examples() |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
413 rows = range(i.start,i.stop,i.step) |
41 | 414 # or a list of indices |
415 elif type(i) is list: | |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
416 rows = i |
41 | 417 if rows is not None: |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
418 examples = [self[row] for row in rows] |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
419 fields_values = zip(*examples) |
45
a5c70dc42972
Test functions for dataset.py
bengioy@grenat.iro.umontreal.ca
parents:
44
diff
changeset
|
420 return MinibatchDataSet( |
41 | 421 Example(self.fieldNames(),[ self.valuesVStack(fieldname,field_values) |
422 for fieldname,field_values | |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
423 in zip(self.fieldNames(),fields_values)]), |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
424 self.valuesVStack,self.valuesHStack) |
41 | 425 # else check for a fieldname |
426 if self.hasFields(i): | |
427 return self.minibatches(fieldnames=[i],minibatch_size=len(self),n_batches=1,offset=0).next()[0] | |
428 # else we are trying to access a property of the dataset | |
429 assert i in self.__dict__ # else it means we are trying to access a non-existing property | |
430 return self.__dict__[i] | |
22
b6b36f65664f
Created virtual sub-classes of DataSet: {Finite{Length,Width},Sliceable}DataSet,
bengioy@esprit.iro.umontreal.ca
parents:
20
diff
changeset
|
431 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
432 def valuesHStack(self,fieldnames,fieldvalues): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
433 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
434 Return a value that corresponds to concatenating (horizontally) several field values. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
435 This can be useful to merge some fields. The implementation of this operation is likely |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
436 to involve a copy of the original values. When the values are numpy arrays, the |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
437 result should be numpy.hstack(values). If it makes sense, this operation should |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
438 work as well when each value corresponds to multiple examples in a minibatch |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
439 e.g. if each value is a Ni-vector and a minibatch of length L is a LxNi matrix, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
440 then the result should be a Lx(N1+N2+..) matrix equal to numpy.hstack(values). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
441 The default is to use numpy.hstack for numpy.ndarray values, and a list |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
442 pointing to the original values for other data types. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
443 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
444 all_numpy=True |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
445 for value in fieldvalues: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
446 if not type(value) is numpy.ndarray: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
447 all_numpy=False |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
448 if all_numpy: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
449 return numpy.hstack(fieldvalues) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
450 # the default implementation of horizontal stacking is to put values in a list |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
451 return fieldvalues |
26
672fe4b23032
Fixed dataset errors so that _test_dataset.py works again.
bengioy@grenat.iro.umontreal.ca
parents:
23
diff
changeset
|
452 |
672fe4b23032
Fixed dataset errors so that _test_dataset.py works again.
bengioy@grenat.iro.umontreal.ca
parents:
23
diff
changeset
|
453 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
454 def valuesVStack(self,fieldname,values): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
455 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
456 Return a value that corresponds to concatenating (vertically) several values of the |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
457 same field. This can be important to build a minibatch out of individual examples. This |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
458 is likely to involve a copy of the original values. When the values are numpy arrays, the |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
459 result should be numpy.vstack(values). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
460 The default is to use numpy.vstack for numpy.ndarray values, and a list |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
461 pointing to the original values for other data types. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
462 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
463 all_numpy=True |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
464 for value in values: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
465 if not type(value) is numpy.ndarray: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
466 all_numpy=False |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
467 if all_numpy: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
468 return numpy.vstack(values) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
469 # the default implementation of vertical stacking is to put values in a list |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
470 return values |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
471 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
472 def __or__(self,other): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
473 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
474 dataset1 | dataset2 returns a dataset whose list of fields is the concatenation of the list of |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
475 fields of the argument datasets. This only works if they all have the same length. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
476 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
477 return HStackedDataSet(self,other) |
3
378b68d5c4ad
Added first (untested) version of ArrayDataSet
bengioy@bengiomac.local
parents:
2
diff
changeset
|
478 |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
479 def __and__(self,other): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
480 """ |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
481 dataset1 & dataset2 is a dataset that concatenates the examples from the argument datasets |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
482 (and whose length is the sum of the length of the argument datasets). This only |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
483 works if they all have the same fields. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
484 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
485 return VStackedDataSet(self,other) |
23
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
486 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
487 def hstack(datasets): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
488 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
489 hstack(dataset1,dataset2,...) returns dataset1 | datataset2 | ... |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
490 which is a dataset whose fields list is the concatenation of the fields |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
491 of the individual datasets. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
492 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
493 assert len(datasets)>0 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
494 if len(datasets)==1: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
495 return datasets[0] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
496 return HStackedDataSet(datasets) |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
497 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
498 def vstack(datasets): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
499 """ |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
500 vstack(dataset1,dataset2,...) returns dataset1 & datataset2 & ... |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
501 which is a dataset which iterates first over the examples of dataset1, then |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
502 over those of dataset2, etc. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
503 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
504 assert len(datasets)>0 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
505 if len(datasets)==1: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
506 return datasets[0] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
507 return VStackedDataSet(datasets) |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
508 |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
509 class FieldsSubsetDataSet(DataSet): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
510 """ |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
511 A sub-class of DataSet that selects a subset of the fields. |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
512 """ |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
513 def __init__(self,src,fieldnames): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
514 self.src=src |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
515 self.fieldnames=fieldnames |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
516 assert src.hasFields(*fieldnames) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
517 self.valuesHStack = src.valuesHStack |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
518 self.valuesVStack = src.valuesVStack |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
519 |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
520 def __len__(self): return len(self.src) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
521 |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
522 def fieldNames(self): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
523 return self.fieldnames |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
524 |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
525 def __iter__(self): |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
526 class FieldsSubsetIterator(object): |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
527 def __init__(self,ds): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
528 self.ds=ds |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
529 self.src_iter=ds.src.__iter__() |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
530 self.example=None |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
531 def __iter__(self): return self |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
532 def next(self): |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
533 complete_example = self.src_iter.next() |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
534 if self.example: |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
535 self.example._values=[complete_example[field] |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
536 for field in self.ds.fieldnames] |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
537 else: |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
538 self.example=Example(self.ds.fieldnames, |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
539 [complete_example[field] for field in self.ds.fieldnames]) |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
540 return self.example |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
541 return FieldsSubsetIterator(self) |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
542 |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
543 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
544 assert self.hasFields(*fieldnames) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
545 return self.src.minibatches_nowrap(fieldnames,minibatch_size,n_batches,offset) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
546 def __getitem__(self,i): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
547 return FieldsSubsetDataSet(self.src[i],self.fieldnames) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
548 |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
549 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
550 class DataSetFields(LookupList): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
551 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
552 Although a DataSet iterates over examples (like rows of a matrix), an associated |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
553 DataSetFields iterates over fields (like columns of a matrix), and can be understood |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
554 as a transpose of the associated dataset. |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
555 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
556 To iterate over fields, one can do |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
557 * for fields in dataset.fields() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
558 * for fields in dataset(field1,field2,...).fields() to select a subset of fields |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
559 * for fields in dataset.fields(field1,field2,...) to select a subset of fields |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
560 and each of these fields is iterable over the examples: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
561 * for field_examples in dataset.fields(): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
562 for example_value in field_examples: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
563 ... |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
564 but when the dataset is a stream (unbounded length), it is not recommanded to do |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
565 such things because the underlying dataset may refuse to access the different fields in |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
566 an unsynchronized ways. Hence the fields() method is illegal for streams, by default. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
567 The result of fields() is a DataSetFields object, which iterates over fields, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
568 and whose elements are iterable over examples. A DataSetFields object can |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
569 be turned back into a DataSet with its examples() method: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
570 dataset2 = dataset1.fields().examples() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
571 and dataset2 should behave exactly like dataset1 (in fact by default dataset2==dataset1). |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
572 |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
573 DataSetFields can be concatenated vertically or horizontally. To be consistent with |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
574 the syntax used for DataSets, the | concatenates the fields and the & concatenates |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
575 the examples. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
576 """ |
74
b4159cbdc06b
Fixed errors raised by test_dataset
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
73
diff
changeset
|
577 def __init__(self,dataset,fieldnames): |
65
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
578 original_dataset=dataset |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
579 if not fieldnames: |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
580 fieldnames=dataset.fieldNames() |
65
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
581 elif not fieldnames==dataset.fieldNames(): |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
582 dataset = FieldsSubsetDataSet(dataset,fieldnames) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
583 assert dataset.hasFields(*fieldnames) |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
584 self.dataset=dataset |
66
dde1fb1b63ba
fixed test and removed print
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
65
diff
changeset
|
585 |
64
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
586 if isinstance(dataset,MinibatchDataSet): |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
587 LookupList.__init__(self,fieldnames,list(dataset._fields)) |
65
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
588 elif isinstance(original_dataset,MinibatchDataSet): |
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
589 LookupList.__init__(self,fieldnames, |
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
590 [original_dataset._fields[field] |
d48eba49a2f4
fixed the infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
64
diff
changeset
|
591 for field in fieldnames]) |
64
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
592 else: |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
593 minibatch_iterator = dataset.minibatches(fieldnames, |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
594 minibatch_size=len(dataset), |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
595 n_batches=1) |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
596 minibatch=minibatch_iterator.next() |
863da25a60f1
trying to fix infinite loop
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
62
diff
changeset
|
597 LookupList.__init__(self,fieldnames,minibatch) |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
598 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
599 def examples(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
600 return self.dataset |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
601 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
602 def __or__(self,other): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
603 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
604 fields1 | fields2 is a DataSetFields that whose list of examples is the concatenation |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
605 of the list of examples of DataSetFields fields1 and fields2. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
606 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
607 return (self.examples() + other.examples()).fields() |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
608 |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
609 def __and__(self,other): |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
610 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
611 fields1 + fields2 is a DataSetFields that whose list of fields is the concatenation |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
612 of the fields of DataSetFields fields1 and fields2. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
613 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
614 return (self.examples() | other.examples()).fields() |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
615 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
616 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
617 class MinibatchDataSet(DataSet): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
618 """ |
85 | 619 Turn a LookupList of same-length (iterable) fields into an example-iterable dataset. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
620 Each element of the lookup-list should be an iterable and sliceable, all of the same length. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
621 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
622 def __init__(self,fields_lookuplist,values_vstack=DataSet().valuesVStack, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
623 values_hstack=DataSet().valuesHStack): |
17
759d17112b23
more comments, looping ArrayDataSet iterator, bugfixes to lookup_list, more tests
bergstrj@iro.umontreal.ca
diff
changeset
|
624 """ |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
625 The user can (and generally should) also provide values_vstack(fieldname,fieldvalues) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
626 and a values_hstack(fieldnames,fieldvalues) functions behaving with the same |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
627 semantics as the DataSet methods of the same name (but without the self argument). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
628 """ |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
629 self._fields=fields_lookuplist |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
630 assert len(fields_lookuplist)>0 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
631 self.length=len(fields_lookuplist[0]) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
632 for field in fields_lookuplist[1:]: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
633 assert self.length==len(field) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
634 self.values_vstack=values_vstack |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
635 self.values_hstack=values_hstack |
3
378b68d5c4ad
Added first (untested) version of ArrayDataSet
bengioy@bengiomac.local
parents:
2
diff
changeset
|
636 |
378b68d5c4ad
Added first (untested) version of ArrayDataSet
bengioy@bengiomac.local
parents:
2
diff
changeset
|
637 def __len__(self): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
638 return self.length |
28
541a273bc89f
Removed __array__ method from dataset, whose
bengioy@grenat.iro.umontreal.ca
parents:
26
diff
changeset
|
639 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
640 def __getitem__(self,i): |
80 | 641 if type(i) in (slice,list): |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
642 return DataSetFields(MinibatchDataSet( |
80 | 643 Example(self._fields.keys(),[field[i] for field in self._fields])),self.fieldNames()) |
644 if type(i) is int: | |
85 | 645 return Example(self._fields.keys(),[field[i] for field in self._fields]) |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
646 if self.hasFields(i): |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
647 return self._fields[i] |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
648 assert i in self.__dict__ # else it means we are trying to access a non-existing property |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
649 return self.__dict__[i] |
11
be128b9127c8
Debugged (to the extent of my tests) the new version of dataset
bengioy@esprit.iro.umontreal.ca
parents:
9
diff
changeset
|
650 |
29
46c5c90019c2
Changed apply_function so that it propagates methods of the source.
bengioy@grenat.iro.umontreal.ca
parents:
28
diff
changeset
|
651 def fieldNames(self): |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
652 return self._fields.keys() |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
653 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
654 def hasFields(self,*fieldnames): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
655 for fieldname in fieldnames: |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
656 if fieldname not in self._fields.keys(): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
657 return False |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
658 return True |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
659 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
660 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
661 class Iterator(object): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
662 def __init__(self,ds): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
663 self.ds=ds |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
664 self.next_example=offset |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
665 assert minibatch_size > 0 |
41 | 666 if offset+minibatch_size > ds.length: |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
667 raise NotImplementedError() |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
668 def __iter__(self): |
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
669 return self |
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
670 def next(self): |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
671 upper = self.next_example+minibatch_size |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
672 assert upper<=self.ds.length |
61
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
673 minibatch = Example(self.ds._fields.keys(), |
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
674 [field[self.next_example:upper] |
a8b70a9117ad
bugfix: in MinibatchDataSet renamed the class variable fields to _fields as parent class have a function called field.
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
60
diff
changeset
|
675 for field in self.ds._fields]) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
676 self.next_example+=minibatch_size |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
677 return minibatch |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
678 |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
679 return Iterator(self) |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
680 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
681 def valuesVStack(self,fieldname,fieldvalues): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
682 return self.values_vstack(fieldname,fieldvalues) |
20
266c68cb6136
Minor editions, plus adding untested ApplyFunctionDataset for GradientLearner in the works.
bengioy@bengiomac.local
parents:
19
diff
changeset
|
683 |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
684 def valuesHStack(self,fieldnames,fieldvalues): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
685 return self.values_hstack(fieldnames,fieldvalues) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
686 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
687 class HStackedDataSet(DataSet): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
688 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
689 A DataSet that wraps several datasets and shows a view that includes all their fields, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
690 i.e. whose list of fields is the concatenation of their lists of fields. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
691 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
692 If a field name is found in more than one of the datasets, then either an error is |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
693 raised or the fields are renamed (either by prefixing the __name__ attribute |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
694 of the dataset + ".", if it exists, or by suffixing the dataset index in the argument list). |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
695 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
696 TODO: automatically detect a chain of stacked datasets due to A | B | C | D ... |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
697 """ |
41 | 698 def __init__(self,datasets,accept_nonunique_names=False,description=None,field_types=None): |
699 DataSet.__init__(self,description,field_types) | |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
700 self.datasets=datasets |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
701 self.accept_nonunique_names=accept_nonunique_names |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
702 self.fieldname2dataset={} |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
703 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
704 def rename_field(fieldname,dataset,i): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
705 if hasattr(dataset,"__name__"): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
706 return dataset.__name__ + "." + fieldname |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
707 return fieldname+"."+str(i) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
708 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
709 # make sure all datasets have the same length and unique field names |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
710 self.length=None |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
711 names_to_change=[] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
712 for i in xrange(len(datasets)): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
713 dataset = datasets[i] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
714 length=len(dataset) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
715 if self.length: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
716 assert self.length==length |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
717 else: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
718 self.length=length |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
719 for fieldname in dataset.fieldNames(): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
720 if fieldname in self.fieldname2dataset: # name conflict! |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
721 if accept_nonunique_names: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
722 fieldname=rename_field(fieldname,dataset,i) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
723 names2change.append((fieldname,i)) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
724 else: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
725 raise ValueError("Incompatible datasets: non-unique field name = "+fieldname) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
726 self.fieldname2dataset[fieldname]=i |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
727 for fieldname,i in names_to_change: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
728 del self.fieldname2dataset[fieldname] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
729 self.fieldname2dataset[rename_field(fieldname,self.datasets[i],i)]=i |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
730 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
731 def hasFields(self,*fieldnames): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
732 for fieldname in fieldnames: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
733 if not fieldname in self.fieldname2dataset: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
734 return False |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
735 return True |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
736 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
737 def fieldNames(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
738 return self.fieldname2dataset.keys() |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
739 |
41 | 740 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
741 |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
742 class HStackedIterator(object): |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
743 def __init__(self,hsds,iterators): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
744 self.hsds=hsds |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
745 self.iterators=iterators |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
746 def __iter__(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
747 return self |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
748 def next(self): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
749 # concatenate all the fields of the minibatches |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
750 return reduce(LookupList.__add__,[iterator.next() for iterator in self.iterators]) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
751 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
752 assert self.hasfields(fieldnames) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
753 # find out which underlying datasets are necessary to service the required fields |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
754 # and construct corresponding minibatch iterators |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
755 if fieldnames: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
756 datasets=set([]) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
757 fields_in_dataset=dict([(dataset,[]) for dataset in datasets]) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
758 for fieldname in fieldnames: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
759 dataset=self.datasets[self.fieldnames2dataset[fieldname]] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
760 datasets.add(dataset) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
761 fields_in_dataset[dataset].append(fieldname) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
762 datasets=list(datasets) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
763 iterators=[dataset.minibatches(fields_in_dataset[dataset],minibatch_size,n_batches,offset) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
764 for dataset in datasets] |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
765 else: |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
766 datasets=self.datasets |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
767 iterators=[dataset.minibatches(None,minibatch_size,n_batches,offset) for dataset in datasets] |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
768 return HStackedIterator(self,iterators) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
769 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
770 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
771 def valuesVStack(self,fieldname,fieldvalues): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
772 return self.datasets[self.fieldname2dataset[fieldname]].valuesVStack(fieldname,fieldvalues) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
773 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
774 def valuesHStack(self,fieldnames,fieldvalues): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
775 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
776 We will use the sub-dataset associated with the first fieldname in the fieldnames list |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
777 to do the work, hoping that it can cope with the other values (i.e. won't care |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
778 about the incompatible fieldnames). Hence this heuristic will always work if |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
779 all the fieldnames are of the same sub-dataset. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
780 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
781 return self.datasets[self.fieldname2dataset[fieldnames[0]]].valuesHStack(fieldnames,fieldvalues) |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
782 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
783 class VStackedDataSet(DataSet): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
784 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
785 A DataSet that wraps several datasets and shows a view that includes all their examples, |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
786 in the order provided. This clearly assumes that they all have the same field names |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
787 and all (except possibly the last one) are of finite length. |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
788 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
789 TODO: automatically detect a chain of stacked datasets due to A + B + C + D ... |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
790 """ |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
791 def __init__(self,datasets): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
792 self.datasets=datasets |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
793 self.length=0 |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
794 self.index2dataset={} |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
795 assert len(datasets)>0 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
796 fieldnames = datasets[-1].fieldNames() |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
797 self.datasets_start_row=[] |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
798 # We use this map from row index to dataset index for constant-time random access of examples, |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
799 # to avoid having to search for the appropriate dataset each time and slice is asked for. |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
800 for dataset,k in enumerate(datasets[0:-1]): |
48
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
801 assert dataset.is_unbounded() # All VStacked datasets (except possibly the last) must be bounded (have a length). |
b6730f9a336d
Fixing MinibatchDataSet getitem
bengioy@grenat.iro.umontreal.ca
parents:
46
diff
changeset
|
802 L=len(dataset) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
803 for i in xrange(L): |
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
804 self.index2dataset[self.length+i]=k |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
805 self.datasets_start_row.append(self.length) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
806 self.length+=L |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
807 assert dataset.fieldNames()==fieldnames |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
808 self.datasets_start_row.append(self.length) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
809 self.length+=len(datasets[-1]) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
810 # If length is very large, we should use a more memory-efficient mechanism |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
811 # that does not store all indices |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
812 if self.length>1000000: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
813 # 1 million entries would require about 60 meg for the index2dataset map |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
814 # TODO |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
815 print "A more efficient mechanism for index2dataset should be implemented" |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
816 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
817 def __len__(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
818 return self.length |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
819 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
820 def fieldNames(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
821 return self.datasets[0].fieldNames() |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
822 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
823 def hasFields(self,*fieldnames): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
824 return self.datasets[0].hasFields(*fieldnames) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
825 |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
826 def locate_row(self,row): |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
827 """Return (dataset_index, row_within_dataset) for global row number""" |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
828 dataset_index = self.index2dataset[row] |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
829 row_within_dataset = self.datasets_start_row[dataset_index] |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
830 return dataset_index, row_within_dataset |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
831 |
41 | 832 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
833 |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
834 class VStackedIterator(object): |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
835 def __init__(self,vsds): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
836 self.vsds=vsds |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
837 self.next_row=offset |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
838 self.next_dataset_index,self.next_dataset_row=self.vsds.locate_row(offset) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
839 self.current_iterator,self.n_left_at_the_end_of_ds,self.n_left_in_mb= \ |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
840 self.next_iterator(vsds.datasets[0],offset,n_batches) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
841 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
842 def next_iterator(self,dataset,starting_offset,batches_left): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
843 L=len(dataset) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
844 ds_nbatches = (L-starting_offset)/minibatch_size |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
845 if batches_left is not None: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
846 ds_nbatches = max(batches_left,ds_nbatches) |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
847 if minibatch_size>L: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
848 ds_minibatch_size=L |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
849 n_left_in_mb=minibatch_size-L |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
850 ds_nbatches=1 |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
851 else: |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
852 n_left_in_mb=0 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
853 return dataset.minibatches(fieldnames,minibatch_size,ds_nbatches,starting_offset), \ |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
854 L-(starting_offset+ds_nbatches*minibatch_size), n_left_in_mb |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
855 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
856 def move_to_next_dataset(self): |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
857 if self.n_left_at_the_end_of_ds>0: |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
858 self.current_iterator,self.n_left_at_the_end_of_ds,self.n_left_in_mb= \ |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
859 self.next_iterator(vsds.datasets[self.next_dataset_index], |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
860 self.n_left_at_the_end_of_ds,1) |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
861 else: |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
862 self.next_dataset_index +=1 |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
863 if self.next_dataset_index==len(self.vsds.datasets): |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
864 self.next_dataset_index = 0 |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
865 self.current_iterator,self.n_left_at_the_end_of_ds,self.n_left_in_mb= \ |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
866 self.next_iterator(vsds.datasets[self.next_dataset_index],starting_offset,n_batches) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
867 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
868 def __iter__(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
869 return self |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
870 |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
871 def next(self): |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
872 dataset=self.vsds.datasets[self.next_dataset_index] |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
873 mb = self.next_iterator.next() |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
874 if self.n_left_in_mb: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
875 extra_mb = [] |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
876 while self.n_left_in_mb>0: |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
877 self.move_to_next_dataset() |
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
878 extra_mb.append(self.next_iterator.next()) |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
879 mb = Example(fieldnames, |
40
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
880 [dataset.valuesVStack(name, |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
881 [mb[name]]+[b[name] for b in extra_mb]) |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
882 for name in fieldnames]) |
88fd1cce08b9
replaced infinity for length by raise UnboundedDataSet and use & instead of + to concatenate datasets
bengioy@esprit.iro.umontreal.ca
parents:
39
diff
changeset
|
883 |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
884 self.next_row+=minibatch_size |
38
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
885 self.next_dataset_row+=minibatch_size |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
886 if self.next_row+minibatch_size>len(dataset): |
d637ad8f7352
Finished first untested version of VStackedDataset
bengioy@esprit.iro.umontreal.ca
parents:
37
diff
changeset
|
887 self.move_to_next_dataset() |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
888 return examples |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
889 return VStackedIterator(self) |
37
73c4212ba5b3
Factored the minibatch-writing code into an iterator class inside DataSet
bengioy@esprit.iro.umontreal.ca
parents:
36
diff
changeset
|
890 |
41 | 891 class ArrayFieldsDataSet(DataSet): |
892 """ | |
893 Virtual super-class of datasets whose field values are numpy array, | |
894 thus defining valuesHStack and valuesVStack for sub-classes. | |
895 """ | |
896 def __init__(self,description=None,field_types=None): | |
897 DataSet.__init__(self,description,field_types) | |
898 def valuesHStack(self,fieldnames,fieldvalues): | |
899 """Concatenate field values horizontally, e.g. two vectors | |
900 become a longer vector, two matrices become a wider matrix, etc.""" | |
901 return numpy.hstack(fieldvalues) | |
902 def valuesVStack(self,fieldname,values): | |
903 """Concatenate field values vertically, e.g. two vectors | |
904 become a two-row matrix, two matrices become a longer matrix, etc.""" | |
905 return numpy.vstack(values) | |
906 | |
907 class ArrayDataSet(ArrayFieldsDataSet): | |
908 """ | |
909 An ArrayDataSet stores the fields as groups of columns in a numpy tensor, | |
910 whose first axis iterates over examples, second axis determines fields. | |
911 If the underlying array is N-dimensional (has N axes), then the field | |
912 values are (N-2)-dimensional objects (i.e. ordinary numbers if N=2). | |
913 """ | |
914 | |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
915 def __init__(self, data_array, fields_columns): |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
916 """ |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
917 Construct an ArrayDataSet from the underlying numpy array (data) and |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
918 a map (fields_columns) from fieldnames to field columns. The columns of a field are specified |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
919 using the standard arguments for indexing/slicing: integer for a column index, |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
920 slice for an interval of columns (with possible stride), or iterable of column indices. |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
921 """ |
41 | 922 self.data=data_array |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
923 self.fields_columns=fields_columns |
41 | 924 |
925 # check consistency and complete slices definitions | |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
926 for fieldname, fieldcolumns in self.fields_columns.items(): |
41 | 927 if type(fieldcolumns) is int: |
928 assert fieldcolumns>=0 and fieldcolumns<data_array.shape[1] | |
929 elif type(fieldcolumns) is slice: | |
930 start,step=None,None | |
931 if not fieldcolumns.start: | |
932 start=0 | |
933 if not fieldcolumns.step: | |
934 step=1 | |
935 if start or step: | |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
936 self.fields_columns[fieldname]=slice(start,fieldcolumns.stop,step) |
41 | 937 elif hasattr(fieldcolumns,"__iter__"): # something like a list |
938 for i in fieldcolumns: | |
939 assert i>=0 and i<data_array.shape[1] | |
940 | |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
941 def fieldNames(self): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
942 return self.fields_columns.keys() |
41 | 943 |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
944 def __len__(self): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
945 return len(self.data) |
41 | 946 |
80 | 947 def __getitem__(self,key): |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
948 """More efficient implementation than the default __getitem__""" |
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
949 fieldnames=self.fields_columns.keys() |
80 | 950 if type(key) is int: |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
951 return Example(fieldnames, |
80 | 952 [self.data[key,self.fields_columns[f]] for f in fieldnames]) |
953 if type(key) is slice: | |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
954 return MinibatchDataSet(Example(fieldnames, |
80 | 955 [self.data[key,self.fields_columns[f]] for f in fieldnames])) |
956 if type(key) is list: | |
957 for i in range(len(key)): | |
958 if self.hasFields(key[i]): | |
959 key[i]=self.fields_columns[key[i]] | |
960 return MinibatchDataSet(Example(fieldnames, | |
88
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
961 #we must separate differently for list as numpy |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
962 # don't support self.data[[i1,...],[i2,...]] |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
963 # when their is more then two i1 and i2 |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
964 [self.data[key,:][:,self.fields_columns[f]] |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
965 if isinstance(self.fields_columns[f],list) else |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
966 self.data[key,self.fields_columns[f]] |
6749d18e11c8
bugfix as numpy numpy don't support self.data[[i1,...],[i2,...]] when their is more then two i1 and i2
Frederic Bastien <bastienf@iro.umontreal.ca>
parents:
85
diff
changeset
|
967 for f in fieldnames]), |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
968 self.valuesVStack,self.valuesHStack) |
80 | 969 |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
970 # else check for a fieldname |
80 | 971 if self.hasFields(key): |
972 return self.data[self.fields_columns[key],:] | |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
973 # else we are trying to access a property of the dataset |
80 | 974 assert key in self.__dict__ # else it means we are trying to access a non-existing property |
975 return self.__dict__[key] | |
55
66619ce44497
Efficient implementation of getitem for ArrayDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
48
diff
changeset
|
976 |
41 | 977 |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
978 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
979 class ArrayDataSetIterator(object): |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
980 def __init__(self,dataset,fieldnames,minibatch_size,n_batches,offset): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
981 if fieldnames is None: fieldnames = dataset.fieldNames() |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
982 # store the resulting minibatch in a lookup-list of values |
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
983 self.minibatch = LookupList(fieldnames,[0]*len(fieldnames)) |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
984 self.dataset=dataset |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
985 self.minibatch_size=minibatch_size |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
986 assert offset>=0 and offset<len(dataset.data) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
987 assert offset+minibatch_size<=len(dataset.data) |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
988 self.current=offset |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
989 def __iter__(self): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
990 return self |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
991 def next(self): |
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
992 sub_data = self.dataset.data[self.current:self.current+self.minibatch_size] |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
993 self.minibatch._values = [sub_data[:,self.dataset.fields_columns[f]] for f in self.minibatch._names] |
43
e92244f30116
Corrected iterator logic errors
bengioy@grenat.iro.umontreal.ca
parents:
42
diff
changeset
|
994 self.current+=self.minibatch_size |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
995 return self.minibatch |
42
9b68774fcc6b
Testing basic functionality and removing obvious bugs
bengioy@grenat.iro.umontreal.ca
parents:
41
diff
changeset
|
996 |
44
5a85fda9b19b
Fixed some more iterator bugs
bengioy@grenat.iro.umontreal.ca
parents:
43
diff
changeset
|
997 return ArrayDataSetIterator(self,fieldnames,minibatch_size,n_batches,offset) |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
998 |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
999 |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1000 class CachedDataSet(DataSet): |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1001 """ |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1002 Wrap a dataset whose values are computationally expensive to obtain |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1003 (e.g. because they involve some computation, or disk access), |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1004 so that repeated accesses to the same example are done cheaply, |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1005 by caching every example value that has been accessed at least once. |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1006 |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1007 Optionally, for finite-length dataset, all the values can be computed |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1008 (and cached) upon construction of the CachedDataSet, rather at the |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1009 first access. |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1010 |
77
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1011 @todo when cache_all_upon_construction create mini-batches that are as |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1012 large as possible but not so large as to fill up memory. |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1013 |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1014 @todo add disk-buffering capability, so that when the cache becomes too |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1015 big for memory, we cache things on disk, trying to keep in memory only |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1016 the record most likely to be accessed next. |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1017 """ |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1018 def __init__(self,source_dataset,cache_all_upon_construction=False): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1019 self.source_dataset=source_dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1020 self.cache_all_upon_construction=cache_all_upon_construction |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1021 if cache_all_upon_construction: |
77
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1022 # this potentially brings all the source examples |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1023 # into memory at once, which may be too much |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1024 # the work could possibly be done by minibatches |
1e2bb5bad636
toying with different ways to implement learners
bengioy@bengiomac.local
parents:
74
diff
changeset
|
1025 # that are as large as possible but no more than what memory allows. |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1026 self.cached_examples = zip(*source_dataset.minibatches(minibatch_size=len(source_dataset)).__iter__().next()) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1027 else: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1028 self.cached_examples = [] |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1029 |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1030 self.fieldNames = source_dataset.fieldNames |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1031 self.hasFields = source_dataset.hasFields |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1032 self.valuesHStack = source_dataset.valuesHStack |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1033 self.valuesVStack = source_dataset.valuesVStack |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1034 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1035 def __len__(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1036 return len(self.source_dataset) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1037 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1038 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1039 class CacheIterator(object): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1040 def __init__(self,dataset): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1041 self.dataset=dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1042 self.current=offset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1043 def __iter__(self): return self |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1044 def next(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1045 upper = self.current+minibatch_size |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1046 cache_len = len(self.dataset.cached_examples) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1047 if upper>=cache_len: # whole minibatch is not already in cache |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1048 # cache everything from current length to upper |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1049 for example in self.dataset.source_dataset[cache_len:upper]: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1050 self.dataset.cached_examples.append(example) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1051 all_fields_minibatch = Example(self.dataset.fieldNames(), |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1052 self.dataset.cached_examples[self.current:self.current+minibatch_size]) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1053 if self.dataset.fieldNames()==fieldnames: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1054 return all_fields_minibatch |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1055 return Example(fieldnames,[all_fields_minibatch[name] for name in fieldnames]) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1056 return CacheIterator(self) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1057 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1058 |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1059 class ApplyFunctionDataSet(DataSet): |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1060 """ |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1061 A dataset that contains as fields the results of applying a given function |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1062 example-wise or minibatch-wise to all the fields of an input dataset. |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1063 The output of the function should be an iterable (e.g. a list or a LookupList) |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1064 over the resulting values. |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1065 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1066 In minibatch mode, the function is expected to work on minibatches (takes |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1067 a minibatch in input and returns a minibatch in output). More precisely, |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1068 it means that each element of the input or output list should be iterable |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1069 and indexable over the individual example values (typically these |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1070 elements will be numpy arrays). All of the elements in the input and |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1071 output lists should have the same length, which is the length of the |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1072 minibatch. |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1073 |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1074 The function is applied each time an example or a minibatch is accessed. |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1075 To avoid re-doing computation, wrap this dataset inside a CachedDataSet. |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1076 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1077 If the values_{h,v}stack functions are not provided, then |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1078 the input_dataset.values{H,V}Stack functions are used by default. |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1079 """ |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1080 def __init__(self,input_dataset,function,output_names,minibatch_mode=True, |
74
b4159cbdc06b
Fixed errors raised by test_dataset
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
73
diff
changeset
|
1081 values_hstack=None,values_vstack=None, |
73
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1082 description=None,fieldtypes=None): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1083 """ |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1084 Constructor takes an input dataset that has as many fields as the function |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1085 expects as inputs. The resulting dataset has as many fields as the function |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1086 produces as outputs, and that should correspond to the number of output names |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1087 (provided in a list). |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1088 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1089 Note that the expected semantics of the function differs in minibatch mode |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1090 (it takes minibatches of inputs and produces minibatches of outputs, as |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1091 documented in the class comment). |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1092 """ |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1093 self.input_dataset=input_dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1094 self.function=function |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1095 self.output_names=output_names |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1096 self.minibatch_mode=minibatch_mode |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1097 DataSet.__init__(description,fieldtypes) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1098 self.valuesHStack = values_hstack if values_hstack else input_dataset.valuesHStack |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1099 self.valuesVStack = values_vstack if values_vstack else input_dataset.valuesVStack |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1100 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1101 def __len__(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1102 return len(self.input_dataset) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1103 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1104 def fieldnames(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1105 return self.output_names |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1106 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1107 def minibatches_nowrap(self,fieldnames,minibatch_size,n_batches,offset): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1108 class ApplyFunctionIterator(object): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1109 def __init__(self,output_dataset): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1110 self.input_dataset=output_dataset.input_dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1111 self.output_dataset=output_dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1112 self.input_iterator=input_dataset.minibatches(minibatch_size=minibatch_size, |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1113 n_batches=n_batches,offset=offset).__iter__() |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1114 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1115 def __iter__(self): return self |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1116 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1117 def next(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1118 function_inputs = self.input_iterator.next() |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1119 all_output_names = self.output_dataset.output_names |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1120 if self.output_dataset.minibatch_mode: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1121 function_outputs = self.output_dataset.function(function_inputs) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1122 else: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1123 input_examples = zip(*function_inputs) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1124 output_examples = [self.output_dataset.function(input_example) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1125 for input_example in input_examples] |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1126 function_outputs = [self.output_dataset.valuesVStack(name,values) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1127 for name,values in zip(all_output_names, |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1128 zip(*output_examples))] |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1129 all_outputs = Example(all_output_names,function_outputs) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1130 if fieldnames==all_output_names: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1131 return all_outputs |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1132 return Example(fieldnames,[all_outputs[name] for name in fieldnames]) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1133 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1134 return ApplyFunctionIterator(self.input_dataset,self) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1135 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1136 def __iter__(self): # only implemented for increased efficiency |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1137 class ApplyFunctionSingleExampleIterator(object): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1138 def __init__(self,output_dataset): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1139 self.current=0 |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1140 self.output_dataset=output_dataset |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1141 self.input_iterator=output_dataset.input_dataset.__iter__() |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1142 def __iter__(self): return self |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1143 def next(self): |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1144 function_inputs = self.input_iterator.next() |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1145 if self.output_dataset.minibatch_mode: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1146 function_outputs = [output[0] for output in self.output_dataset.function(function_inputs)] |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1147 else: |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1148 function_outputs = self.output_dataset.function(function_inputs) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1149 return Example(self.output_dataset.output_names,function_outputs) |
69f97aad3faf
Coded untested ApplyFunctionDataSet and CacheDataSet
bengioy@bengiomac.local
parents:
72
diff
changeset
|
1150 return ApplyFunctionSingleExampleIterator(self) |
57
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1151 |
1aabd2e2bb5f
Added empty classes with doc: CachedDataSet and ApplyFunctionDataSet
Yoshua Bengio <bengioy@iro.umontreal.ca>
parents:
56
diff
changeset
|
1152 |
23
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1153 def supervised_learning_dataset(src_dataset,input_fields,target_fields,weight_field=None): |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1154 """ |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1155 Wraps an arbitrary DataSet into one for supervised learning tasks by forcing the |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1156 user to define a set of fields as the 'input' field and a set of fields |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1157 as the 'target' field. Optionally, a single weight_field can also be defined. |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1158 """ |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1159 args = ((input_fields,'input'),(output_fields,'target')) |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1160 if weight_field: args+=(([weight_field],'weight')) |
36
438440ba0627
Rewriting dataset.py completely
bengioy@zircon.iro.umontreal.ca
parents:
29
diff
changeset
|
1161 return src_dataset.merge_fields(*args) |
23
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1162 |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1163 |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1164 |
526e192b0699
Working on ApplyFunctionDataSet, added constraint that
bengioy@esprit.iro.umontreal.ca
parents:
22
diff
changeset
|
1165 |