pylearn: doc/v2_planning/learner.txt annotate

annotate doc/v2_planning/learner.txt @ 1474:a57f4839a9d8

merge

author	James Bergstra <bergstrj@iro.umontreal.ca>
date	Wed, 18 May 2011 10:52:42 -0400
parents	0e12ea6ba661
children

rev	line source
1041 38cc6e075d9b PV added to learner committee Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1038 diff changeset	1
38cc6e075d9b PV added to learner committee Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1038 diff changeset	2 Comittee: AB, PL, GM, IG, RP, NB, PV
1167 7a8dcf87d780 Rename learn_meeting.py to API_learner.txt Pascal Lamblin <lamblinp@iro.umontreal.ca> parents: 1059 diff changeset	3 Leader: PL
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	4
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	5 Discussion of Function Specification for Learner Types
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	6 ======================================================
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	7
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	8 In its most abstract form, a learner is an object with the
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	9 following semantics:
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	10
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	11 * A learner has named hyper-parameters that control how it learns (these can be viewed
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	12 as options of the constructor, or might be set directly by a user)
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	13
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	14 * A learner also has an internal state that depends on what it has learned.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	15
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	16 * A learner reads and produces data, so the definition of learner is
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	17 intimately linked to the definition of dataset (and task).
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	18
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	19 * A learner has one or more 'train' or 'adapt' functions by which
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	20 it is given a sample of data (typically either the whole training set, or
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	21 a mini-batch, which contains as a special case a single 'example'). Learners
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	22 interface with datasets in order to obtain data. These functions cause the
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	23 learner to change its internal state and take advantage to some extent
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	24 of the data provided. The 'train' function should take charge of
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	25 completely exploiting the dataset, as specified per the hyper-parameters,
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	26 so that it would typically be called only once. An 'adapt' function
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	27 is meant for learners that can operate in an 'online' setting where
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	28 data continually arrive and the control loop (when to stop) is to
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	29 be managed outside of it. For most intents and purposes, the
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	30 'train' function could also handle the 'online' case by providing
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	31 the controlled iterations over the dataset (which would then be
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	32 seen as a stream of examples).
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	33
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	34 * learner.train(dataset)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	35 * learner.adapt(data)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	36
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	37 * Different types of learners can then exploit their internal state
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	38 in order to perform various computations after training is completed,
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	39 or in the middle of training, e.g.,
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	40
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	41 * y=learner.predict(x)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	42 for learners that see (x,y) pairs during training and predict y given x,
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	43 or for learners that see only x's and learn a transformation of it (i.e. feature extraction).
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	44 Here and below, x and y are tensor-like objects whose first index iterates
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	45 over particular examples in a batch or minibatch of examples.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	46
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	47 * p=learner.probability(examples)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	48 p=learner.log_probability(examples)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	49 for learners that can estimate probability density or probability functions,
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	50 note that example could be a pair (x,y) for learners that expect each example
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	51 to represent such a pair. The second form is provided in case the example
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	52 is high-dimensional and computations in the log-domain are numerically preferable.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	53 The first dimension of examples or of x and y is an index over a minibatch or a dataset.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	54
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	55 * p=learner.free_energy(x)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	56 for learners that can estimate a log unnormalized probability; the output has the same length as the input.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	57
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	58 * c=learner.costs(examples)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	59 returns a matrix of costs (one row per example, i.e., again the output has the same length
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	60 as the input), the first column of which represents the cost whose expectation
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	61 we wish to minimize over new samples from the unknown underlying data distribution.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	62
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	63
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	64 Some learners may be able to handle x's and y's that contain missing values.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	65
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	66 * For convenience, some of these operations could be bundled, e.g.
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	67
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	68 * [prediction,costs] = learner.predict_and_adapt((x,y))
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	69
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	70 * Some learners could include in their internal state not only what they
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	71 have learned but some information about recently seen examples that conditions
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	72 the expected distribution of upcoming examples. In that case, they might
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	73 be used, e.g. in an online setting as follows:
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	74
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	75 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	76
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	77 for (x,y) in data_stream:
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	78 [prediction,costs]=learner.predict((x,y))
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	79 accumulate_statistics(prediction,costs)
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	80
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	81 * In some cases, each example is itself a (possibly variable-size) sequence
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	82 or other variable-size object (e.g. an image, or a video)
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	83
f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	84
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	85
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	86
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	87
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	88
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	89
1002 f82093bf4405 adding learner.txt and dataset.txt in v2_planning/ Yoshua Bengio <bengioy@iro.umontreal.ca> parents: diff changeset	90
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	91
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	92
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	93 James's idea for Learner Interface
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	94 ===================================
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	95
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	96 Theory:
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	97 -------
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	98
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	99 Think about the unfolding of a learning algorithm as exploring a path in a vast
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	100 directed graph.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	101
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	102 There are some source nodes, which are potential initial conditions for the
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	103 learning algorithm.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	104
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	105 At any node, there are a number of outgoing labeled edges that represent
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	106 distinct directions of exploration: like "allocate a model with N hidden units",
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	107 or "set the l1 weight decay on such-and-such units to 0.1" or "adapt for T
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	108 iterations" or "refresh the GPU dataset memory with the next batch of data".
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	109
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	110 Not all nodes have the same outgoing edge labels. The dataset, model, and
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	111 optimization algorithm implementations may each have their various
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	112 hyper-parameters with various restrictions on what values they can take, and
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	113 when they can be changed.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	114
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	115 Every move in this graph incurs some storage and computational expense, and
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	116 explores the graph.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	117
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	118 Learners typically engage in goal-directed exploration of this graph - for
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	119 example to find the node with the best validation-set performance given a
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	120 certain computational budget. We might often be interested in the best node
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	121 found.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	122
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	123 The predict(), log_probability(), free_energy() etc correspond to costs that we
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	124 can measure at any particular node (at some computational expense) to see how we
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	125 are doing in our exploration.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	126
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	127 Many semantically distinct components come into the definition of this graph:
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	128 the model (e.g. DAA) the dataset (e.g. an online one), the inference and
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	129 learning strategy. I'm not sure what to call this graph than an 'experiment
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	130 graph'... so I'll go with that for now.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	131
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	132
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	133
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	134
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	135
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	136 Use Cases
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	137 ----------
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	138
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	139 Early stopping
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	140 ~~~~~~~~~~~~~~
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	141
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	142 Early stopping can be implemented as a learner that progresses along a
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	143 particular kind of edge (e.g. "train more") until a stopping criterion (in terms
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	144 of a cost computed from nodes along the path) is met.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	145
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	146
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	147 Grid Search
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	148 ~~~~~~~~~~~
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	149
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	150 Grid search is a learner policy that can be implemented in an experiment graph
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	151 where all paths have the form:
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	152
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	153 ( "set param 0 to X",
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	154 "set param 1 to Y",
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	155 ... ,
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	156 "set param N to Z",
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	157 adapt,
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	158 [early stop...],
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	159 test)
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	160
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	161 It would explore all paths of this form and then return the best node.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	162
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	163
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	164 Stagewise learning of DBNs combined with early stopping and grid search
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	165 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	166
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	167 This would be a learner that is effective for experiment graphs that reflect the
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	168 greedy-stagewise optimization of DBNs.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	169
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	170
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	171 Boosting
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	172 ~~~~~~~~
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	173
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	174 Given an ExperimentGraph that permits re-weighting of examples, it is
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	175 straightforward to write a meta-ExperimentGraph around it that implements AdaBoost.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	176 A meta-meta-ExperimentGraph around that that does early-stopping would complete
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	177 the picture and make a useful boosting implementation.
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	178
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	179
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	180 Using External Hyper-Parameter Optimization Software
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	181 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	182 TODO: use-case - show how we could use the optimizer from
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	183 http://www.cs.ubc.ca/labs/beta/Projects/ParamILS/
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	184
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	185
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	186 Implementation Details / API
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	187 ----------------------------
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	188
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	189 Learner
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	190 ~~~~~~~
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	191 An object that allows us to explore the graph discussed above. Specifically, it represents
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	192 an explored node in that graph.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	193
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	194 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	195
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	196 def active_instructions()
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	197 """ Return a list/set of Instruction instances (see below) that the Learner is prepared
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	198 to handle.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	199 """
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	200
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	201 def copy(), deepcopy()
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	202 """ Learners should be serializable """
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	203
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	204
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	205 To make the implementation easier, I found it was helpful to introduce a string-valued
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	206 `fsa_state` member attribute and associate methods to these states. That made it
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	207 syntactically easy to build relatively complex finite-state transition graphs to describe
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	208 which instructions were active at which times in the life-cycle of a learner.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	209
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	210
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	211 Instruction
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	212 ~~~~~~~~~~~
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	213 An object that represents a potential edge in the graph discussed above. It is an
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	214 operation that a learner can perform.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	215
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	216 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	217
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	218 arg_types
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	219 """a list of Type object (see below) indicating what args are required by execute"""
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	220
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	221 def execute(learner, args, kwargs):
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	222 """ Perform some operation on the learner (follow an edge in the graph discussed above)
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	223 and modify the learner in-place. Calling execute 'moves' the learner from one node in
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	224 the graph along an edge. To have the old learner as well, it must be copied prior to
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	225 calling execute().
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	226 """
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	227
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	228 def expense(learner, args, kwargs, resource_type='CPUtime'):
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	229 """ Return an estimated cost of performing this instruction (calling execute), in time,
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	230 space, number of computers, disk requierement, etc.
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	231 """
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	232
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	233 Type
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	234 ~~~~
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	235 An object that describes a parameter domain for a call to Instruction.execute.
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	236 It is not necessary that a Type specifies exactly which arguments are legal, but it should
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	237 `include` all legal arguments, and exclude as many illegal ones as possible.
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	238
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	239 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	240
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	241 def includes(value):
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	242 """return True if value is a legal argument"""
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	243
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	244
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	245 To make things a bit more practical, there are some Type subclasses like Int, Float, Str,
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	246 ImageDataset, SgdOptimizer, that include additional attributes (e.g. min, max, default) so
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	247 that automatic graph exploration algorithms can generate legal arguments with reasonable
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	248 efficiency.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	249
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	250
38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	251
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	252 The proxy pattern is a powerful way to combine learners. Especially when proxy Learner
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	253 instances also introduce Proxy Instruction classes.
1026 38f799f8b6cd v2_planning - thoughts on learner James Bergstra <bergstrj@iro.umontreal.ca> parents: 1002 diff changeset	254
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	255 For example, it is straightforward to implement a hyper-learner by implementing a Learner with
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	256 another learner (sub-learner) as a member attribute. The hyper-learner makes some
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	257 modifications to the instruction_set() return value of the sub-learner, typically to introduce
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	258 more powerful instructions and hide simpler ones.
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	259
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	260 It is less straightforward, but consistent with the design to implement a Learner that
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	261 encompasses job management. Such a learner would retain the semantics of the
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	262 instruction_set of the sub-learner, but would replace the Instruction objects themselves with
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	263 Instructions that arranged for remote procedure calls (e.g. jobman, multiprocessing, bqtools,
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	264 etc.) Such a learner would replace synchronous instructions (return on completion) with
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	265 asynchronous ones (return after scheduling) and the active instruction set would also change
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	266 asynchronously, but neither of these things is inconsistent with the Learner API.
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	267
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	268
1058 e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	269 TODO - Experiment API?
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	270 ~~~~~~~~~~~~~~~~~~~~~~
1043 3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	271
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	272 I feel like something is missing from the API - and that is an interface to the graph structure
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	273 discussed above. The nodes in this graph are natural places to store meta-information for
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	274 visualization, statistics-gathering etc. But none of the APIs above corresponds to the graph
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	275 itself. In other words, there is no API through which to attach information to nodes. It is
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	276 not good to say that the Learner instance is the node because (a) learner instances change
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	277 during graph exploration and (b) learner instances are big, and we don't want to have to keep a
3f528656855b v2planning learner.txt - updated API recommendation James Bergstra <bergstrj@iro.umontreal.ca> parents: 1041 diff changeset	278 whole saved model just to attach meta-info e.g. validation score. Choosing this API spills
1058 e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	279 over into other committees, so we should get their feedback about how to resolve
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	280 it. Maybe we need an 'Experiment' API to stand for this graph?
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	281
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	282
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	283 TODO: Validation & Monitoring Costs
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	284 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	285
1058 e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	286 Even if we do have the Experiment API as a structure to hang validation and
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	287 monitoring results, what should be the mechanism for extracting those results.
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	288 The Learner API is not right because extracting a monitoring cost doesn't change
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	289 the model, doesn't change the legal instructions/edges etc. Maybe we should use
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	290 a similar mechanism to Instruction, called something like Measurement? Any node
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	291 / learner can report the list of instructions (for moving) and the list of
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	292 measurements (and the cost of computing them too)
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	293
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	294
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	295 TODO - Parameter Distributions
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	296 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1055 bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	297
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	298 YB asks: it seems to me that what we really need from "Type" is not just
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	299 testing that a value is legal, but more practically a function that specifies the
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	300 prior distribution for the hyper-parameter, i.e., how to sample from it,
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	301 and possibly some representation of it that could be used to infer
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	302 a posterior (such as an unnormalized log-density or log-probability).
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	303 Having the min and max and default limits us to the uniform distribution,
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	304 which may not always be appropriate. For example sometimes we'd like
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	305 Gaussian (-infty to infty) or Exponential (0 to infty) or Poisson (non-negative integers).
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	306 For that reason, I think that "Type" is not a very good name.
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	307 How about "Prior" or "Density" or something like that?
bc3f7834db83 added a comment/question about Type Yoshua Bengio <bengioy@iro.umontreal.ca> parents: 1053 diff changeset	308
1058 e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	309 JB replies: I agree that being able to choose (and update) distributions over
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	310 these values is important. I don't think the Type structure is the right place
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	311 to handle it though. The challenge is to allow those distributions to change
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	312 for a variety of reasons - e.g. the sampling distribution on the capacity
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	313 variables is affected by the size of the dataset, it is also affected by
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	314 previous experience in general as well as experiments on that particular
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	315 dataset. I'm not sure that the 'Type' structure is right to deal with this.
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	316 Also, even with a strategy for handling these distributions, I believe a simple
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	317 mechanism for rejecting insane values might be useful.
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	318
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	319 So how should we handle it? Hmmm...
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	320
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	321
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	322 Comments
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	323 ~~~~~~~~
e342de3ae485 v2planning learner - added comments and TODO points James Bergstra <bergstrj@iro.umontreal.ca> parents: 1055 diff changeset	324
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	325 OD asks: (I hope it's ok to leave comments even though I'm not in committee... I'm
1045 d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	326 interested to see how the learner interface is shaping up so I'll be keeping
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	327 an eye on this file)
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	328 I'm wondering what's the benefit of such an API compared to simply defining a
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	329 new method for each instruction. It seems to me that typically, the 'execute'
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	330 method would end up being something like
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	331
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	332 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	333
1045 d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	334 if instruction == 'do_x':
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	335 self.do_x(..)
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	336 elif instruction == 'do_y':
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	337 self.do_y(..)
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	338 ...
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	339
1045 d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	340 so why not directly call do_x / do_y instead?
d57bdd9a9980 learner: Left a comment about James' design Olivier Delalleau <delallea@iro> parents: 1044 diff changeset	341
1046 f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	342
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	343 JB replies: I agree with you, and in the implementation of a Learner I suggest
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	344 using Python decorators to get the best of both worlds:
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	345
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	346 .. code-block:: python
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	347
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	348 class NNet(Learner):
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	349
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	350 ...
1046 f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	351
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	352 @Instruction.new(arg_types=(Float(min=-8, max=-1, default=-4),))
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	353 def set_log_lr(self, log_lr):
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	354 self.lr.value = numpy.exp(log_lr)
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	355
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	356 ...
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	357
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	358 The Learner base class can implement a instruction_set() that walks through the
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	359 methods of 'self' and pick out the ones that have corresponding instructions.
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	360 But anyone can call the method normally. The NNet class can also have methods
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	361 that are not instructions.
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	362
1053 390166ace9e5 learner: Reply to James Olivier Delalleau <delallea@iro> parents: 1052 diff changeset	363 OD replies: Ok thanks. I'm still unsure what is the end goal, but I'll keep
390166ace9e5 learner: Reply to James Olivier Delalleau <delallea@iro> parents: 1052 diff changeset	364 watching while you guys work on it, and hopefully it'll become clearer for me ;)
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	365
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	366 RP asks: James correct me if I'm wrong, but I think each instruction has a execute
1046 f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	367 command. The job of the learner is to traverse the graph and for each edge
f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	368 that it decides to cross to call the execute of that edge. Maybe James has
f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	369 something else in mind, but this was my understanding.
f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	370
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	371 JB replies: close, but let me make a bit of a clarification. The job of a
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	372 Learner is simply to implement the API of a Learner - to list what edges are
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	373 available and to be able to cross them if asked. The code using the Learner
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	374 (client) decides which edges to cross. The client may also be a Learner, but
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	375 maybe not.
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	376
1046 f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	377
f1732269bce8 comment on Olivier's comment Razvan Pascanu <r.pascanu@gmail.com> parents: 1045 diff changeset	378
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	379 Just another view/spin on the same idea (Razvan)
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	380 ================================================
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	381
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	382
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	383 My idea is probably just a spin off from what James wrote. It is an extension
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	384 of what I send on the mailing list some time ago.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	385
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	386 Big Picture
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	387 -----------
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	388
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	389 What do we care about ?
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	390 ~~~~~~~~~~~~~~~~~~~~~~~
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	391
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	392 This is the list of the main points that I have in mind :
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	393
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	394 * Re-usability
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	395 * Extensibility
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	396 * Simplicity or easily readable code ( connected to re-usability )
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	397 * Modularity ( connected to extensibility )
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	398 * Fast to write code ( - sort of comes out of simplicity)
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	399 * Efficient code
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	400
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	401
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	402 Composition
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	403 ~~~~~~~~~~~
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	404
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	405 To me this reads as code generated by composing pieces. Imagine this :
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	406 you start of with something primitive that I will call a "variable", which
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	407 probably is a very unsuitable name. And then you compose those intial
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	408 "variables" or transform them through several "functions". Each such
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	409 "function" hides some logic, that you as the user don't care about.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	410 You can have low-level or micro "functions" and high-level or macro
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	411 "functions", where a high-level function is just a certain compositional
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	412 pattern of low-level "functions". There are several classes of "functions"
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	413 and "variables" that can be inter-changable. This is how modularity is
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	414 obtained, by chainging between functions from a certain class.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	415
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	416 Now when you want to research something, what you do is first select
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	417 the step you want to look into. If you are lucky you can re-write this
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	418 step as certain decomposition of low-level transformations ( there can be
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	419 multiple such decompositions). If not you have to implement such a
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	420 decompositions acording to your needs. Pick the low-level transformations you want
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	421 to change and write new versions that implement your logic.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	422
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	423 I think the code will be easy to read, because it is just applying a fixed
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	424 set of transformations, one after the other. The one who writes the code can
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	425 decide how explicit he wants to write things by switching between high-level
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	426 and low-level functions.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	427
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	428 I think the code this way is re-usable, because you can just take this chain
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	429 of transformation and replace the one you care about, without looking into
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	430 the rest.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	431
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	432 You get this fractal property of the code. Zooming in, you always get just
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	433 a set of functions applied to a set of variables. In the begining those might
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	434 not be there, and you would have to create new "low level" decompositions,
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	435 maybe even new "variables" that get data between those decompositions.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	436
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	437 The thing with variables here, is that I don't want this "functions" to have
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	438 a state. All the information is passed along through these variables. This
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	439 way understanding the graph is easy, debugging it is also easier ( then having
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	440 all these hidden states ..)
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	441
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	442 Note that while doing so we might ( and I strongly think we should) create
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	443 a (symbolic) DAG of operations. ( this is where it becomes what James was saying).
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	444 In such a DAG the "variables" will the nodes and the functions will be edges.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	445 I think having a DAG is useful in many ways (all this are things that one
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	446 might think about implementing in a far future, I'm not proposing to implement
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	447 them unless we want to use them - like the reconstruction ):
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	448
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	449 * there exist the posibility of writing optimizations ( theano style )
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	450 * there exist the posibility to add global view utility functions ( like
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	451 a reconstruction function for SdA - extremely low level here), or global
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	452 view diagnostic tools
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	453 * the posibility of creating a GUI ( where you just create the Graph by
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	454 picking transforms and variables from a list ) or working interactively
0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	455 and then generating code that will reproduce the graph
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	456 * you can view the graph and different granularity levels to understand
1189 0e12ea6ba661 fix many rst syntax error warning. Frederic Bastien <nouiz@nouiz.org> parents: 1167 diff changeset	457 things ( global diagnostics)
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	458
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	459 We should have a taxonomy of possible classes of functions and possible
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	460 classes of variables, but those should not be exclusive. We can work at a high
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	461 level for now, and decompose those high level functions to lower level when
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	462 we need to. We can introduce new classes of functions or intermediate
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	463 variables between those low level functions.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	464
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	465
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	466 Similarities with James' idea
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	467 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	468
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	469 As I said before, this is I think just another view on what James proposed.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	470 The learner in his case is the module that traverses the graph of this
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	471 operations, which makes sense here as well.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	472
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	473 The 'execute' command in his api is just applying a function to some variables in
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	474 my case.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	475
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	476 The learner keeps track of the graph that is formed I think in both cases.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	477
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	478 His view is a bit more general. I see the graph as fully created by the user,
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	479 and the learner just has to go from the start to the end. In his case the
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	480 traversal is conditioned on some policies. I think these ideas can be mixed /
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	481 united. What I would see in my case to have this functionality is something
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	482 similar to the lazy linker for Theano.
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	483
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	484
3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	485
1052 84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	486 JB asks: There is definitely a strong theme of graphs in both suggestions,
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	487 furthermore graphs that have heavy-duty nodes and light-weight edges. But I
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	488 don't necessarily think that we're proposing the same thing. One difference is
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	489 that the graph I talked about would be infinite in most cases of interest, so
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	490 it's not going to be representable by Theano's data structures (even with lazy
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	491 if). Another difference is that the graph I had in mind doesn't feel fractal -
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	492 it would be very common for a graph edge to be atomic. A proxy pattern, such as
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	493 in a hyper-learner would create a notion of being able to zoom in, but other
84f62533e7a8 v2planning learner - reply to comments James Bergstra <bergstrj@iro.umontreal.ca> parents: 1046 diff changeset	494 than that, i'm not sure what you mean.
1044 3b1fd599bafd my first draft of my own views which are close to be just a reformulation of what James proposes Razvan Pascanu <r.pascanu@gmail.com> parents: 1043 diff changeset	495
1056 19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	496 RP replies: I've been thinking about my idea a bit and yes, it might be
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	497 quite different from what James has in mind, though there are plently of common
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	498 elements. I might have exagerated a bit with the zooming in, so in some cases
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	499 you will end up with atomic edges, though my hope is that is not most of the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	500 edges.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	501
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	502 I think I should go into mode details when answering this question because
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	503 I feel I have not explained things sufficiently clear. Note, in many places
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	504 I replaced the word "function" by "transform".
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	505
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	506 Think of the learner as an object that traverses a DAG of steps created by the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	507 user. On this DAG the learner can potentially do a lot of cool stuff, but we
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	508 won't care about that for now. The DAG can be infinite in principle, and what
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	509 the learner does is just to go on the path described by the user ( and here
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	510 described is not through heuristics like in James case, but by giving the list
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	511 of edges it needs to follow). A potential cool thing the learner can do is to
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	512 regard the path given by the user as a suggestion ( or some form of heuristic)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	513 and try to improve it. This would be much closer to what James has in mind,
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	514 and I definetely think is a cool way to go about it.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	515
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	516 Now this path in the graph is given by the user by composing subgraphs or
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	517 adding nodes to the graph. Or (expressing this in a more simple way) by applying
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	518 functions to variables. Any such function will introduce an edge ( or a subgraph) that
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	519 will connect the vertices corresponding to the input variables to the vertices
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	520 corresponding to the output variables. The variables store the state of the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	521 learner. These functions are state-less, I think if you would give them states
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	522 you will make this approach really ugly (I might be wrong).
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	523 The variables would contain informations required by the function, like
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	524 number of layers, on how many cores to run, cluster configurations, and so on.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	525
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	526 Now about the zooming part, that James asked. I might have exagerated a bit,
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	527 is not that you can zoom in on any part infinitely. You will end up with
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	528 things that are atomic. The idea is that any such "transformation" or edge
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	529 has the potential to be split up in several "transformations". This offers
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	530 (in my view) a way of solving the time constraints of our project. We can
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	531 start by difining a coarse division in segments. For now we can have
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	532 a structure transform that makes a list of parameters into a deep
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	533 network of some type, then a learner transform that adds SGD + pre-training
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	534 on top of network, and then early stopper on top of that, and then a
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	535 run_on_cluster on that.We would probably want something more finely grained
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	536 even from the start .. this is just to prove my point. When any of us
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	537 starts experimenting with a certain sub-step of this process ( like the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	538 structure) we will split that transform into several ( like ones that create
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	539 a layer and so on) that make sense for that case, and then start working on
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	540 the low level transform that we cares ( like the layer) introducing new
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	541 versions of it. I think we can not find a universal split that will cover
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	542 all of our cases, so I think we should allow different such splits. The one
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	543 who researches should look at what low-level transforms are available and use
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	544 those if they make sense, if not he would have to create a different split.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	545 Creating a different split might involve a lot of work and taking care of
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	546 several issues so it should be done with care.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	547
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	548 I'll give an example from where I started thinking this way. Let say we want
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	549 to do the SdA with auxiliary inputs that encourages separation of the features
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	550 in the hidden layer that Yoshua was saying ( I had an attempt
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	551 at it some time ago for speech but I never eneded up finishing that project).
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	552
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	553 You start up with something like :
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	554
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	555 learner = Learner()
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	556 # This will create the learner that will traverse our graph. We might
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	557 # want it to be a function ``execute``, I just randomly picked this option.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	558 #I have no preference of this detail for now .. this is mostly work in progress
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	559
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	560 data = someSpeechData(path = 'some path')
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	561 # This is such a transform that will generate from the string representing the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	562 # path a dataset variable ( that will contain all informations you need to
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	563 # access data). This will probably be the object the datasets comittee will
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	564 # provide. Note, you might need to provide more information then the path, but
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	565 # you can easily see how to do that. All these stuff start from simple
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	566 # variables like path, batch size and so on and return a complex heavy duty
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	567 # variable (node).
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	568
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	569
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	570 model = earlyStopping(pretrain(SdA(layers = [524, 500, 500,27], noise = [0.1,0.1]),data, epochs = 10), data)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	571 # This is a composition of two transforms. The SdA transform starts from the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	572 # info about layers and corruption /noise for each layer and construct a SdA.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	573 # This is a high level transform, so it will take care of defining all
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	574 # details, like pre-training, defining the cost and so on. Note that maybe it will
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	575 # require some more parameters .. you can assume that for anything else there
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	576 # is a default value that the SdA will use. earlyStopping is yet another
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	577 # transform that takes a model ( that we know how to train ) and some data,
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	578 # and does early stoppign on it. For bravity I did not provide all the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	579 # information required like patience and so on. The SdA only knows how to do a
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	580 # step of training. Same holds for pretrain. It will loop over the layers of
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	581 # SdA and will train each one.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	582
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	583 steps = cluster(model, getPropertiesAndRanges(model), n_jobs = 20, cluster_info = getClusterInfo())
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	584 # This will lunch the wanted jobs. getPropertiesAndRanges will get from a
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	585 # model all knobs that need to be turn, and their ranges and will uniformly
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	586 # sample from them in each jobs. getCluterInfo will return a variable
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	587 # containing informations about the cluster ( I added this for simplicity, it
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	588 # should probably be replaced with something like username, password,
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	589 # clusterpath or whatever).
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	590
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	591 learner.execute(steps)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	592 # As an option, each of this output variables could contain the entire graph
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	593 # until that point. We could also have this in a different way .. this is
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	594 # adhoc at the moment
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	595
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	596
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	597 Now this is a coarse vanila SdA which is not what we wanted. We do not have a
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	598 way of incorporating our auxiliary information in this. So what we have to do
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	599 is split/change the SdA transform. We would re-write it as :
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	600
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	601
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	602 arch = SdA(layers = [524, 500, 500, 27], noise = [0.1,0.1])
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	603 model = earlyStopping(pretrain(arch,data,epochs = 10)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	604 ...
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	605
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	606 And then re-write things like :
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	607
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	608 arch = SGD( cross_entropy( logreg( DAAlayer( [DAAlayer([524,500],0.1),500],0.1))))
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	609
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	610
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	611 We would re-write the DAAlayer as :
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	612
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	613 layer0 = DAAlayer([524,500],0.1)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	614 layer1 = cross_entropy(reconstruct( tanh(dotW_b( layer0,500)),noise = 0.1))
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	615
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	616 At this point of detail, we can start inserting our new stuff in as follows :
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	617
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	618
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	619 input = empty_layer(600)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	620 # empty layer is a wrapper ; if I would to write dotW_b(200,500) which means
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	621 # go from a layer of 200 units to a one of 500 by multiplying with a matrix
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	622 # and adding a bias, what I would mean is dotW_b( empty_layer(200), 500).
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	623 # an implementation of empty_layer could be just theano.tensor.vector()
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	624 # where we add the size tag ( we will need it later)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	625
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	626
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	627 hidden0_mfcc = dotW_b(input[0:524],100)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	628 hidden0_noise = dotW_b(input[0:560],50)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	629 hidden0_speakerID = dotW_b(join(input[0:524], input[560:600]),50)
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	630 hidden0 = tanh(join( layer0_mfcc, layer0_noise, layer0_speakerID))
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	631 layer0 = cross_entropy( reconstruct( hidden0, noise = 0.1))
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	632
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	633 and so on. Hopefully you got what I mean by spliting a transform, or zooming
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	634 in. When doing all this we did not change anything about the early stopping or
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	635 lunching jobs on the cluster. In the same manner, if one would like to look
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	636 into how jobs are send to the cluster, it could just expand that part. Note
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	637 that if we wanted to do something else we might have split the DAA
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	638 differently.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	639
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	640 The key of this approach is to identify such low level units that can be
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	641 shared by 90% of our architectures, and the splits that make most sense
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	642 from a functional point of view that will cover the main points where people
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	643 will like to change things. This will ensure that almost all the time we have
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	644 the wanted low-level bits that we want to write our code into, and most of the
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	645 time we will only work on one of that bit. There will definetely be cases when
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	646 whatever we have will not be sufficient or convinient. In that case some
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	647 effort has to be invested by the user to create a different decomposition of
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	648 the problem in the elements he need.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	649
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	650 I've been thinking about this a bit, and it definetely works in for deep
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	651 networks and theano ( the approach was inspired by theano). From what James
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	652 said, I think that other stuff might be possible to incorporate, at least as
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	653 atomic transforms if not in any other way.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	654
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	655 TODO: one has to give some thought of this low-level transform, to find a
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	656 suitable set of them ( and variables) so that would end up most of the time
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	657 re-using things and not creating new things.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	658
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	659 NOTES: there are some other implementation details missing of what this state
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	660 variables should contain. I did not want to clutter this with what tricks
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	661 could be used to get this transparent interface. I have a few of them in mind
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	662 though..
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	663 there is a lot of hardcoded values in this example. Usually each transform
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	664 that takes an input should "know" which of these inputs are tunable and mark
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	665 them as such. The order of the input in this example is important as well.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	666 This can be easily solved at the expense of a few more lines of code that
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	667 I did not want to write.
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	668
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	669
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	670
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	671
19033ef1636d some more details on my approach Razvan Pascanu <r.pascanu@gmail.com> parents: 1055 diff changeset	672

Mercurial > pylearn

annotate doc/v2_planning/learner.txt @ 1474:a57f4839a9d8