pylearn: doc/v2_planning/datalearn.txt annotate

annotate doc/v2_planning/datalearn.txt @ 1390:746ebceeb46f

added comments to hmc code (old outstanding changes)

author	gdesjardins
date	Mon, 20 Dec 2010 18:08:04 -0500
parents	decee534c78d
children	e8fc563dad74

rev	line source
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	1 DataLearn: How to plug Datasets & Learner together?
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	2 ===================================================
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	3
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	4 Participants
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	5 ------------
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	6 - Yoshua
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	7 - Razvan
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	8 - Olivier D [leader]
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	9
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	10 High-Level Objectives
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	11 ---------------------
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	12
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	13 * Simple ML experiments should be simple to write
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	14 * More complex / advanced scenarios should be possible without being forced
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	15 to work "outside" of this framework
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	16 * Computations should be optimized whenever possible
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	17 * Existing code (in any language) should be "wrappable" within this
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	18 framework
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	19 * It should be possible to replace [parts of] this framework with C++ code
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	20
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	21 Theano-Like Data Flow
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	22 ---------------------
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	23
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	24 We want to rely on Theano to be able to take advantage of its efficient
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	25 computations. The general idea is that if we chain multiple processing
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	26 elements (think e.g. of a feature selection step followed by a PCA projection,
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	27 then a rescaling within a fixed bounded interval), the overall transformation
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	28 from input to output data can be represented by a Theano symbolic graph. When
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	29 one wants to access the actual numeric data, a function is compiled so as to
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	30 do these computations efficiently.
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	31
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	32 We discussed some specific API options for datasets and learners, which will
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	33 be added to this file in the future, but a core question that we feel should
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	34 be addressed first is how this Theano-based implementation could be achieved
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	35 exactly. For this purpose, in the following, let us assume that a dataset is
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	36 simply a matrix whose rows represent individual samples, and columns
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	37 individual features. How to handle field names, non-tensor-like data, etc. is
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	38 a very important topic that is not yet discussed in this file.
ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	39
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	40 A question we did not discuss much is to which extent the architecture could
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	41 be "theanified", i.e. whether a whole experiment could be defined as a Theano
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	42 graph on which high level optimizations could be made possible, while also
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	43 relying on Theano to "run" the graph. The other option is to use a different
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	44 mechanism, with underlying Theano graphs being built wherever possible to link
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	45 the various components of an experiment together.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	46
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	47 For now, let us consider the latter option, where each dataset contains a
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	48 pointer to a Theano variable that represents the data stored in this dataset.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	49 One issue with this approach is illustrated by the following example. Imagine
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	50 we want to iterate on samples in a dataset and do something with their numeric
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	51 value. We would want the code to be as close as possible to:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	52
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	53 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	54
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	55 for sample in dataset:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	56 do_something_with(sample.numeric_value())
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	57
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	58 A naive implementation of the sample API could be (assuming each sample also
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	59 contains a ``variable`` member which is the variable representing this
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	60 sample's data):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	61
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	62 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	63
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	64 def numeric_value(self):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	65 if self.function is None:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	66 # Compile function to output the numeric value stored in this
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	67 # sample's variable.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	68 self.function = theano.function([], self.variable)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	69 return self.function()
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	70
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	71 However, this is not a good idea, because it would trigger a new function
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	72 compilation for each sample. Instead, we would want something like this:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	73
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	74 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	75
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	76 def numeric_value(self):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	77 if self.function_storage[0] is None:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	78 # Compile function to output the numeric value stored in this
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	79 # sample's variable. This function takes as input the index of
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	80 # the sample in the dataset, and is shared among all samples.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	81 self.function_storage[0] = theano.function(
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	82 [self.symbolic_index], self.variable)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	83 return self.function(self.numeric_index)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	84
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	85 In the code above, we assume that all samples created by the action of
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	86 iterating over the dataset share the same ``function_storage``,
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	87 ``symbolic_index`` and ``variable``: the first time we try to access the numeric
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	88 value of some sample, a function is compiled, that takes as input the index,
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	89 and outputs the variable. The only difference between samples is thus that
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	90 they are given a different numeric value for the index (``numeric_index``).
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	91
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	92 Another way to obtain the same result is to actually let the user take care of
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	93 compiling the function. It would allow the user to really control what is
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	94 being compiled, at the cost of having to write more code:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	95
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	96 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	97
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	98 symbolic_index = dataset.get_index() # Or just theano.tensor.iscalar()
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	99 get_sample = theano.function([symbolic_index],
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	100 dataset[symbolic_index].variable)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	101 for numeric_index in xrange(len(dataset))
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	102 do_something_with(get_sample(numeric_index))
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	103
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	104 James comments: this is how I have written the last couple of projects, it's
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	105 slightly verbose but it's clear and efficient.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	106
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	107 The code above may also be simplified by providing helper functions. In the
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	108 example above, such a function could allow us to iterate on the numeric values
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	109 of samples in a dataset while taking care of compiling the appropriate Theano
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	110 function. See Discussion: Helper Functions below.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	111
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	112 Note that although the above example focused on how to iterate over a dataset,
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	113 it can be cast into a more generic problem, where some data (either dataset or
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	114 sample) is the result of some transformation applied to other data, which is
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	115 parameterized by parameters p1, p2, ..., pN (in the above example, we were
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	116 considering a sample that was obtained by taking the p1-th element in a
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	117 dataset). If we use different values for a subset Q of the parameters but keep
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	118 other parameters fixed, we would probably want to compile a single function
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	119 that takes as input all parameters in Q, while other parameters are fixed. It
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	120 may be nice to try and get the best of both worlds, letting the user take
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	121 control on what is being compiled, while leaving the option of using a default
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	122 sensible behavior for those who do not want to worry about it. Whether this is
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	123 possible / desirable is still to-be-determined.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	124
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	125 What About Learners?
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	126 --------------------
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	127
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	128 The discussion above only mentioned datasets, but not learners. The learning
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	129 part of a learner is not a main concern (currently). What matters most w.r.t.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	130 what was discussed above is how a learner takes as input a dataset and outputs
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	131 another dataset that can be used with the dataset API.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	132
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	133 A Learner may be able to compute various things. For instance, a Neural
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	134 Network may output a ``prediction`` vector (whose elements correspond to
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	135 estimated probabilities of each class in a classification task), as well as a
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	136 ``cost`` vector (whose elements correspond to the penalized NLL, the NLL alone
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	137 and the classification error). We would want to be able to build a dataset
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	138 that contains some of these quantities computed on each sample in the input
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	139 dataset.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	140
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	141 The Neural Network code would then look something like this:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	142
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	143 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	144
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	145 class NeuralNetwork(Learner):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	146
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	147 # The decorator below is reponsible for turning a function that
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	148 # takes a symbolic sample as input, and outputs a Theano variable,
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	149 # into a function that can also be applied on numeric sample data,
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	150 # or symbolic datasets.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	151 # Other approaches than a decorator are possible (e.g. using
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	152 # different function names).
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	153 @datalearn(..)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	154 def compute_prediction(self, sample):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	155 return softmax(theano.tensor.dot(self.weights, sample.input))
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	156
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	157 @datalearn(..)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	158 def compute_nll(self, sample):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	159 return - log(self.compute_prediction(sample)[sample.target])
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	160
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	161 @datalearn(..)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	162 def compute_penalized_nll(self, sample):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	163 return (self.compute_nll(self, sample) +
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	164 theano.tensor.sum(self.weights**2))
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	165
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	166 @datalearn(..)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	167 def compute_class_error(self, sample):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	168 probabilities = self.compute_prediction(sample)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	169 predicted_class = theano.tensor.argmax(probabilities)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	170 return predicted_class != sample.target
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	171
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	172 @datalearn(..)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	173 def compute_cost(self, sample):
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	174 return theano.tensor.concatenate([
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	175 self.compute_penalized_nll(sample),
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	176 self.compute_nll(sample),
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	177 self.compute_class_error(sample),
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	178 ])
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	179
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	180 The ``@datalearn`` decorator would allow such a Learner to be used e.g. like
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	181 this:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	182
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	183 .. code-block:: python
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	184
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	185 nnet = NeuralNetwork()
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	186 # Symbolic dataset that represents the output on symbolic input data.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	187 predict_dataset = nnet.compute_prediction(dataset)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	188 for sample in dataset:
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	189 # Symbolic sample that represents the output on a single symbolic
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	190 # input sample.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	191 predict_sample = nnet.compute_prediction(sample)
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	192 # Numeric prediction.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	193 predict_numeric = nnet.compute_prediction({'input': numpy.zeros(10)})
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	194 # Combining multiple symbolic outputs.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	195 multiple_fields_dataset = ConcatDataSet([
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	196 nnet.compute_prediction(dataset),
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	197 nnet.compute_cost(dataset),
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	198 ])
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	199
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	200 In the code above, if one wants to obtain the numeric value of an element of
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	201 ``multiple_fields_dataset``, the Theano function being compiled should be able
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	202 to optimize computations so that the simultaneous computation of
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	203 ``prediction`` and ``cost`` is done efficiently.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	204
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	205 Discussion: Are Datasets Variables / Ops?
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	206 -----------------------------------------
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	207
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	208 OD wonders: Should datasets directly be Theano Variables, or should they be a
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	209 different object subclass containing a Theano Variable? The advantage of the
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	210 former option would be that they would fit directly within the Theano
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	211 framework, which may allow high level optimizations on data transformations.
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	212 However, we would lose the ability to combine Theano expressions coded in
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	213 individual datasets into a single graph. Currently, I instead considered that
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	214 a dataset has a member that is a Theano variable, and this variable represents
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	215 the data stored in the dataset. The same is done for individual data samples.
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	216
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	217 James asks: Why would a Theano graph in which some nodes represent datasets give
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	218 up the ability to combine Theano expressions coded in individual datasets?
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	219 Firstly, if you want to use Theano expressions and compiled functions to
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	220 implement the perform() method of an Op, you can do that. Secondly, you can
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	221 just include those 'expressions coded in individual datasets' into the overall
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	222 graph.
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	223
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	224 OD replies to James: What I had in mind is you would be forced to compile your
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	225 own function inside the perform() method of an Op. This seemed like a
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	226 potential problem to me because it would prevent Theano from seeing the whole
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	227 fine-grained graph and do optimizations across multiple dataset
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	228 transformations (there may also be additional overhead from calling multiple
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	229 function). But if you are saying it is possible to include 'expressions coded
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	230 in individual datasets' into the overall graph, then I guess this point is
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	231 moot. Would this be achieved with an optimization that replaces the dataset
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	232 node with its internal graph?
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	233
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	234 Razvan comments: 1) Having Theano expressions inside the perform of a Theano
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	235 Op can lead to issues. I know I had to deal with a few when implementing
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	236 Scan which does exactly this. Well to be fair these issues mostly come into
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	237 play when the inner graph has to interact with the outer graph and most of
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	238 the time they can be solved. I guess all that I'm saying is going that way
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	239 might lead to some head-ache to developers, though I guess some head-ache
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	240 will be involved no matter what
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	241 2) In my view (I'm not sure this is what Olivier was saying) the idea of
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	242 not putting the Dataset into a Variable is to not put the logic related to
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	243 loading data, dividing it into slices when running it on the GPU and so on
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	244 into a theano variable. In my view this logic goes into a DataSet class
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	245 that gives you shared variables, symbolic indices into that shared
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	246 variables, and also numeric indices. When looping through those numeric
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	247 indices, the dataset class can reload parts of the data into the
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	248 shared variable and so on.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	249
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	250 OD replies to Razvan's point 2: I think what you are saying is another concern
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	251 I had, which was the fact it may be confusing to mix in the same class the
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	252 Variable/Op and DataSet interfaces. I would indeed prefer to keep them
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	253 separate. However, it may be possible to come up with a system that would get
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	254 the best of both worlds (maybe by having the Op/Variable as members of
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	255 Dataset, and just asking the user building a theano graph to use these instead
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	256 of the dataset directly). Note that I'm mixing up Op/Variable here, because
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	257 it's just not clear yet for me which would go where...
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	258
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	259
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	260 Discussion: Implicit / Explicit Function Compilation
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	261 ----------------------------------------------------
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	262
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	263 <Razvan comments>: I assume that ``do_something_with`` is suppose to be some
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	264 numeric function, and dataset in this case is the result of some
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	265 computations on a initial dataset.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	266 I would differentiate the two approaches (1) and (2) as :
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	267 - first of all whatever you can do with (1) you can do with (2)
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	268 - approach (1) hides the fact that you are working with symbolic graphs.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	269 You apply functions to datasets, and when you want to see values a
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	270 function is compiled under the hood and those values are computed for
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	271 you. In approach (2) the fact that you deal with a symbolic graph is
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	272 explicit because you have to manually compile your functions.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	273 - approach (1) needs to use this function_storage trick shared between
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	274 certain nodes of the graph to reduce the number of compilation while in
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	275 approach (2) we don't need to deal with the complexity of lazy
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	276 compilation
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	277
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	278 OD comments: Well, to be fair, it means we put the burden of dealing with the
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	279 complexity of lazy compilation on the user (it's up to him to make sure he
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	280 compiles only one function).
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	281
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	282 - approach (1) needs a replace function if you want to change the dataset.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	283 What you would do, is once you have a "computational graph" or pipeline
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	284 or whatever you call it, say ``graph``, to change the input you would do
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	285 graph.replace({ init_data_X: new_data_X}), In approach (2) the init_data_X
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	286 and new_data_X is the ``dataset`` so you would compile two different
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	287 functions. Well I would re-write (2) -- to make the above more clear --
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	288 as :
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	289
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	290 .. code-block:: python
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	291
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	292 symbolic_index = theano.tensor.iscalar()
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	293 get_sample1 = theano.function( [symbolic_index],
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	294 graph( dataset[symbolic_index] ).variable)
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	295 for numeric_index in xrange(len(dataset)):
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	296 do_something_with(get_sample(numeric_index))
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	297
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	298 get_sample2 = theano.function( [symbolic_index],
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	299 graph( new_dataset[symbolic_index] ).variable)
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	300 ## Note: the dataset was replaced with new_dataset
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	301 for numeric_index in xrange(len(new_dataset)):
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	302 do_something_with(get_sample2(numeric_index))
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	303
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	304 ######### FOR (1) you write:
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	305
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	306 for datapoint in graph:
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	307 do_something_with( datapoint() )
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	308
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	309 new_graph = graph.replace({dataset:dataset2})
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	310
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	311 for datapoint in new_graph:
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	312 do_something_with(datapoint())
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	313
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	314 OD comments: I don't really understand what is 'graph' in this code (it
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	315 appears in both approaches but is used differently). What I have in mind would
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	316 be more with 'graph' removed in the first approach you describe (#2), and
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	317 graph / new_graph replaced by dataset / new_dataset in the second one (#1).
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	318 You wouldn't need to call some graph.replace method: the graphs compiled for
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	319 iterating on 'dataset' and 'new_dataset' would be entirely separate (using two
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	320 different compiled functions, pretty much like #2).
1363 18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	321
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	322 RP answers: Yes you are right. What I was trying to say is if you have two
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	323 different datasets on which you want to apply the same pre-processing you
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	324 can do that in both approaches. ``graph`` represents the pre-processing
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	325 steps in (2) and the end dataset (after preprocessing) in (1). So the idea
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	326 is that instead of making new_graph from scratch (re-applying all the
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	327 transforms on the original dataset) you can use replace. Or maybe the
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	328 __call__ (that compiles the function if needed) can get a givens dictionary
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	329 ( that replaces datasets or more ). I only gave this argument because I
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	330 thought this will be an issue people will raise. They will say, well in (2)
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	331 the pipeline logic is separated from the data, so you can use the same
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	332 transformation with different data easily, while in (1) you write the
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	333 transformation rooted in a dataset, and if you want same transformation
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	334 for a different dataset you have to re-write everything.
18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	335
1364 01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	336 OD replies: Still not sure I understand. If you have a "graph" function that
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	337 takes a dataset as input and outputs a new dataset, you can use this same
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	338 function with both (1) and (2). With (2) it is:
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	339 theano.function([index], graph(my_dataset)[index].variable)
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	340 while with (1) the same function is compiled implicitly with:
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	341 for sample in graph(my_dataset):
01157763c2d7 Reply to Razvan Olivier Delalleau <delallea@iro> parents: 1363 diff changeset	342 ...
1363 18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	343
1365 049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	344 RP answers: right. I was actually constructing this stupid example in my mind when
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	345 you would do like :
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	346 i1 = f1(data)
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	347 i2 = f2(i1)
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	348 i3 = f3(i2)
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	349 ...
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	350 iN = fN(iN-1)
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	351 and then you would say .. wait I want to do this on new_data as well. Oh no, I
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	352 have to copy the entire block or whatever. That is so annoying. But actually you
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	353 could just write:
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	354
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	355 def my_f(data):
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	356 i1 = f1(data)
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	357 ...
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	358 return iN
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	359
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	360 and then just use that function which is what you pointed out. I agree I'm
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	361 not sure anymore on the point that I was trying to make. Is like if you are
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	362 a lazy programmer, and you write everything without functions, you can
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	363 argue that you like more (2) because you only pass the dataset at the end
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	364 and not at the beginning. But if (1) would have the replace function this
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	365 argument will fail. Though this only stands if you like don't want to make
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	366 a function out of your pipeline that takes the dataset as input, which now
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	367 that I think about it is pretty stupid not to do. Sorry for that.
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	368
049b99f4b323 reply to OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1364 diff changeset	369
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	370 - in approach (1) the initial dataset object (the one that loads the data)
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	371 decides if you will use shared variables and indices to deal with the
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	372 dataset or if you will use ``theano.tensor.matrix`` and not the user( at
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	373 least not without hacking the code). Of course whoever writes that class
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	374 can add a flag to it to switch between behaviours that make sense.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	375 In approach (2) one is not forced to do this
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	376 inside that class by construction, though by convention I would do it.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	377 So if you consider the one who writes that class as a developer than
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	378 in (2) the user can decide/deal with this and not the developer.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	379 Though this is a fine-line -- I would say the user would actually
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	380 write that class as well using some template.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	381 That is to say (2) looks and feels more like working with Theano
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	382 directly,
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	383
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	384 Bottom line, I think (1) puts more stress on the development of the library,
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	385 and hides Theano and some of the complexity for day to day usage.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	386 In (2) everything is a bit more explicit, leaving the impression that you
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	387 have more control over the code, though I strongly feel that whatever can
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	388 be done in (2) can be done in (1). Traditionally I was more inclined
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	389 towards (1) but now I'm not that sure, I think both are equally interesting
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	390 and valid options.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	391 </Razvan comments>
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	392
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	393 Discussion: Fixed Parameters vs. Function Arguments
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	394 ---------------------------------------------------
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	395
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	396 Razvan Comment: I thought about this a bit at the Pylearn level. In my
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	397 original train of thought you would have the distinction between ``hand
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	398 picked parameters`` which I would call hyper-parameter and learned
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	399 parameters. A transformation in this framework (an op if you wish) could
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	400 take as inputs DataSet(s), DataField(s), Parameter(s) (which are the things
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	401 that the learner should adapt) and HyperParameter(s). All hyper-parameters
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	402 will turn into arguments of the compiled function (like the indices of each
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	403 of the dataset objects ) and therefore they can be changed without
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	404 re-compilation. Or in other words this can be easily done by having new
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	405 types of Variables that would represent Parameters and Hyper-parameters.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	406 And as an ending note I would say that there are
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	407 hyper-parameters for which you need to recompile the thenao function and
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	408 can not be just parameters ( so we would have yet another category ?).
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	409
1366 f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	410 Yoshua's comments on RP's comments: I don't understand why we would
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	411 need to create these types. Isn't it just a matter for the programmer
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	412 to decide what are the inputs of the compiled function, and which
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	413 are possibly constant (e.g. holding some hyper-parameters constant
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	414 for a while)?
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	415
1368 ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	416 RP answers: If we opt for this lazy compilation mechanism, the library needs
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	417 to know what to put into a shared, and what to expect as input. The
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	418 programmer should give hints to the library by saying this value will always
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	419 be constant, or this is a hyper-parameter that I might want to change, and
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	420 when I do that I don't want to recompile everything so put it as an
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	421 argument. Even when the compilation is done by the user, it would be helpful
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	422 to have some function that collects all the parameters for you. What I mean
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	423 is that it would be nice to write something like
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	424
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	425 corruption_layer_1 = Parameter ( value = 0.1, name = 'c1')
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	426 # Followed by (many) lines of code
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	427 f = function ( results.inputs()+ results.hyper-params(), result )
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	428
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	429
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	430 where results.hyper-params parses the graph, collects the hyper-parameter
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	431 and returns them as a list of theano.Variables wrappen in theano.In with
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	432 a default value and a name. You could call the function either as
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	433
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	434 f()
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	435 or
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	436 f(c1 = 0.2)
ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	437
1370 5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	438 OD comments: Here is a (hopefully simpler) suggestion to solve this problem.
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	439 Consider any data{set,point} obtained by a transformation of an existing
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	440 data{set,point} with parameters p1, p2, ..., pN. From the point of view of
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	441 theano variables, this is something like x2 = h(x1, p1=v1, ..., pn=vN) where
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	442 x1, x2 are variables and h is an Op. In addition v1 ... vN are also variables
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	443 since they are parameters of the transformation we may want to vary. This is
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	444 not, however, the way the user would build the graph, because being forced to
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	445 use variables for parameters is not user-friendly (IMO). Instead, someone
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	446 would write:
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	447 d2 = t(d1, p1=w1, ..., pn=wN)
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	448 where d1, d2 are data{set,point}s, t is the transformation, and w1 ... wN are
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	449 numeric values of the parameters. Then t would build the piece of graph above,
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	450 so that when you ask d2.numeric_value(), a function computing x2 would be
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	451 compiled, that would take as input variables v1, ... vN.
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	452 Now, the problem is that this may not be fully optimized, since parameters are
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	453 assumed to be varying (so as not to be forced to recompile a different
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	454 function when the user calls t with different parameter values). My suggestion
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	455 is to make this the default behavior, but add an extra argument to t:
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	456 d2 = t(d1, p1=w1, ..., pn=Wn, constants=['p3', 'p5'])
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	457 The line above would do the same, except that the function being compiled
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	458 would use the constant values w3 and w5 for p3 and p5.
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	459 Razvan's example above would be written in a different way as follows:
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	460 def f(c1=0.2):
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	461 return transformK(..(transform2(transform1(input_data,
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	462 corruption_layer_1=c1))))
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	463 With this code you could create various transformed datasets by callling f
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	464 with different values for c1. The first time you call f(c1=0).numeric_value()
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	465 a Theano function is compiled that takes a `corruption_layer_1` input variable
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	466 (whose value is 0 when the function is called by `numeric_value`). If you call
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	467 f().numeric_value(), the same function is re-used (no need to compile it) with
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	468 this input set to 0.2. If on another hand you want to compile a new function
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	469 for each new value of your `corruption_layer_1` parameter, you would instead
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	470 write:
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	471 def f(c1=0.2):
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	472 return transformK(..(transform2(transform1(input_data,
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	473 corruption_layer_1=c1,
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	474 constants=['corruption_layer_1']))))
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	475 This would be one way to have automatic lazy function cache / compilation
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	476 while still letting the user specify for which parameters a new function needs
5785cbac3361 Added a suggestion to solve the problem of Fixed vs. Varying parameters Olivier Delalleau <delallea@iro> parents: 1369 diff changeset	477 to be compiled when their value changes.
1368 ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	478
1371 98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	479 RP comment : What about the same trick that Theano uses, namely, if you want
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	480 a non "default" behaviour you wrap the input in a dictionary. You would
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	481 write tranform1( input_data,
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	482 corruption_layer_1= In(value = c1, fixed = True)) ?
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	483 I started to like this approach of passing extra info about an argument :).
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	484 Other that this it sounds good to me.
98d4232df1d8 comment on OD idea Razvan Pascanu <r.pascanu@gmail.com> parents: 1370 diff changeset	485
1372 decee534c78d Replied to RP Olivier Delalleau <delallea@iro> parents: 1371 diff changeset	486 OD replies: Yes, I guess it would make sense. The more I look at it, the more
decee534c78d Replied to RP Olivier Delalleau <delallea@iro> parents: 1371 diff changeset	487 it seems like it is very close to directly writing a Theano transform on some
decee534c78d Replied to RP Olivier Delalleau <delallea@iro> parents: 1371 diff changeset	488 variables.
decee534c78d Replied to RP Olivier Delalleau <delallea@iro> parents: 1371 diff changeset	489
1368 ad53f73020c2 Answered Yoshua's question Razvan Pascanu <r.pascanu@gmail.com> parents: 1367 diff changeset	490
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	491 Discussion: Helper Functions
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	492 ----------------------------
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	493
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	494 James: Another syntactic option for iterating over datasets is
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	495
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	496 .. code-block:: python
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	497
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	498 for sample in dataset.numeric_iterator(batchsize=10):
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	499 do_something_with(sample)
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	500
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	501 The numeric_iterator would create a symbolic batch index, and compile a single function
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	502 that extracts the corresponding minibatch. The arguments to the
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	503 numeric_iterator function can also specify what compile mode to use, any givens
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	504 you might want to apply, etc.
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	505
1366 f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	506 Yoshua's comment to James' comment: I like that approach.
f945ed016c68 comment by YB Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1365 diff changeset	507
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	508 OD comments: Would there also be some kind of function cache to avoid
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	509 compiling the same function again if we re-iterate on the same dataset with
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	510 the same arguments? Maybe a more generic issue is: would there be a way for
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	511 Theano to be more efficient when re-compiling the same function that was
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	512 already compiled in the same program? (note that I am assuming here it is not
1363 18b2ebec6bca Reply to a comment of OD Razvan Pascanu <r.pascanu@gmail.com> parents: 1362 diff changeset	513 efficient, but I may be wrong).
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	514
1369 f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	515 OD adds: After thinking more about it, this seems very close to my first
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	516 version where a function is automatically compiled "under the hood" when
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	517 iterating on a dataset and accessing the numeric value of a resulting
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	518 sample. The main differences are:
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	519 - In your version, the result is directly a numeric value, while in my version
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	520 one would obtain symbolic samples and would need to call some method to
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	521 obtain their numeric value. I think I like mine a bit better because it
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	522 means you can use the same syntax to e.g. iterate on a dataset, whether you
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	523 are interested in the symbolic representation of samples, or their numeric
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	524 values. On another hand, doing so could be less efficient since you create an
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	525 intermediate representation you may not use. The overhead does not seem much
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	526 to me but I am not sure about that.
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	527 - In your version, you can provide to the function e.g. compile modes /
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	528 givens. This could probably also be done in my version, although it makes it
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	529 more difficult if you want to cache the function to avoid compiling it more
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	530 than once (see next point).
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	531 - (Related to my first comment above) In your version it seems like a new
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	532 function would be compiled every time the user calls e.g.
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	533 'numeric_iterator', while in my version the function would be compiled only
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	534 once. Maybe this can be solved at the Theano level with an efficient
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	535 function cache?
f3a549bd8688 datalearn: Added another comment on James' numeric iterator function Olivier Delalleau <delallea@iro> parents: 1368 diff changeset	536
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	537 Discussion: Dataset as Learner Ouptut
9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	538 -------------------------------------
1357 ffa2932a8cba Added datalearn committee discussion file Olivier Delalleau <delallea@iro> parents: diff changeset	539
1359 5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	540 James asks:
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	541 What's wrong with simply passing the variables corresponding to the dataset to
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	542 the constructor of the learner?
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	543 That seems much more flexible, compact, and clear than the decorator.
5db730bb0e8e comments on datalearn James Bergstra <bergstrj@iro.umontreal.ca> parents: 1357 diff changeset	544
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	545 OD replies: Not sure I understand your idea here. We probably want a learner
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	546 to be able to compute its output on multiple datasets, without having to point
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	547 to these datasets within the learner itself (which seems cumbersome to me).
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	548 The point of the decorators is mostly to turn a single function (that outputs
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	549 a theano variable for the ouptut computed on a single sample) into a function
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	550 that can compute symbolic datasets as well as numeric sample outputs. Those
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	551 could also be instead different functions in the base Learner class if the
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	552 decorator approach is considered ugly / confusing.
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	553
1361 7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	554 Razvan asks: What is predict_sample for ? What is predict_dataset? What I
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	555 guess you mean is that the decorator is used to convert a function that
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	556 takes a theano variable and outputs a theano variable into a class/function
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	557 that takes a DataField/DataSet and outputs a DataField/DataSet. It could
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	558 also register all those different functions, so that the Dataset that
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	559 you get out of (not one of the function) the entire Learner (this Dataset
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	560 is returned by __call__) would contain all those as fields.
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	561 I would use it like this:
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	562
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	563 .. code-block:: python
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	564
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	565 nnet = NeuralNetwork()
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	566 results = nnet(dataset)
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	567 for datapoint in results:
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	568 print datapoint.prediction, datapoint.nll, ...
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	569
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	570 Is this close to what you are suggesting?
7548dc1b163c Some question/suggestions to datalearn Razvan Pascanu <r.pascanu@gmail.com> parents: 1359 diff changeset	571
1362 6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	572 OD: Yes, you guessed right, the decorator's role is to do something different
6b9673d72a41 Datalearn replies / comments Olivier Delalleau <delallea@iro> parents: 1361 diff changeset	573 depending on the input to the function (see my reply to James above).
1367 9474fb4ad109 Refactored datalearn committee file to be easier to read Olivier Delalleau <delallea@iro> parents: 1366 diff changeset	574

Mercurial > pylearn

annotate doc/v2_planning/datalearn.txt @ 1390:746ebceeb46f