ift6266: baseline/conv_mlp/convolutional

annotate baseline/conv_mlp/convolutional_mlp.py @ 352:cfb79f9fd1a4

Ajout d'une fonctionnalite pour pouvoir avoir un taux d'apprentissage decroissant dans le pretrain

author	SylvainPL <sylvain.pannetier.lebeuf@umontreal.ca>
date	Wed, 21 Apr 2010 14:50:59 -0400
parents	d41fe003fade
children

rev	line source
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	1 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	2 This tutorial introduces the LeNet5 neural network architecture using Theano. LeNet5 is a
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	3 convolutional neural network, good for classifying images. This tutorial shows how to build the
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	4 architecture, and comes with all the hyper-parameters you need to reproduce the paper's MNIST
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	5 results.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	6
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	7 The best results are obtained after X iterations of the main program loop, which takes ***
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	8 minutes on my workstation (an Intel Core i7, circa July 2009), and *** minutes on my GPU (an
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	9 NVIDIA GTX 285 graphics processor).
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	10
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	11 This implementation simplifies the model in the following ways:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	12
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	13 - LeNetConvPool doesn't implement location-specific gain and bias parameters
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	14 - LeNetConvPool doesn't implement pooling by average, it implements pooling by max.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	15 - Digit classification is implemented with a logistic regression rather than an RBF network
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	16 - LeNet5 was not fully-connected convolutions at second layer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	17
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	18 References:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	19 - Y. LeCun, L. Bottou, Y. Bengio and P. Haffner: Gradient-Based Learning Applied to Document
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	20 Recognition, Proceedings of the IEEE, 86(11):2278-2324, November 1998.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	21 http://yann.lecun.com/exdb/publis/pdf/lecun-98.pdf
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	22 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	23
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	24 import numpy, theano, cPickle, gzip, time
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	25 import theano.tensor as T
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	26 import theano.sandbox.softsign
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	27 import sys
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	28 import pylearn.datasets.MNIST
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	29 from pylearn.io import filetensor as ft
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	30 from theano.sandbox import conv, downsample
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	31
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	32 from ift6266 import datasets
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	33 import theano,pylearn.version,ift6266
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	34
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	35 class LeNetConvPoolLayer(object):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	36
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	37 def __init__(self, rng, input, filter_shape, image_shape, poolsize=(2,2)):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	38 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	39 Allocate a LeNetConvPoolLayer with shared variable internal parameters.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	40 :type rng: numpy.random.RandomState
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	41 :param rng: a random number generator used to initialize weights
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	42 :type input: theano.tensor.dtensor4
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	43 :param input: symbolic image tensor, of shape image_shape
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	44 :type filter_shape: tuple or list of length 4
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	45 :param filter_shape: (number of filters, num input feature maps,
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	46 filter height,filter width)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	47 :type image_shape: tuple or list of length 4
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	48 :param image_shape: (batch size, num input feature maps,
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	49 image height, image width)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	50 :type poolsize: tuple or list of length 2
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	51 :param poolsize: the downsampling (pooling) factor (#rows,#cols)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	52 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	53 assert image_shape[1]==filter_shape[1]
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	54 self.input = input
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	55
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	56 # initialize weight values: the fan-in of each hidden neuron is
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	57 # restricted by the size of the receptive fields.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	58 fan_in = numpy.prod(filter_shape[1:])
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	59 W_values = numpy.asarray( rng.uniform( \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	60 low = -numpy.sqrt(3./fan_in), \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	61 high = numpy.sqrt(3./fan_in), \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	62 size = filter_shape), dtype = theano.config.floatX)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	63 self.W = theano.shared(value = W_values)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	64
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	65 # the bias is a 1D tensor -- one bias per output feature map
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	66 b_values = numpy.zeros((filter_shape[0],), dtype= theano.config.floatX)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	67 self.b = theano.shared(value= b_values)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	68
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	69 # convolve input feature maps with filters
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	70 conv_out = conv.conv2d(input, self.W,
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	71 filter_shape=filter_shape, image_shape=image_shape)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	72
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	73 # downsample each feature map individually, using maxpooling
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	74 pooled_out = downsample.max_pool2D(conv_out, poolsize, ignore_border=True)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	75
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	76 # add the bias term. Since the bias is a vector (1D array), we first
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	77 # reshape it to a tensor of shape (1,n_filters,1,1). Each bias will thus
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	78 # be broadcasted across mini-batches and feature map width & height
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	79 self.output = T.tanh(pooled_out + self.b.dimshuffle('x', 0, 'x', 'x'))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	80
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	81 # store parameters of this layer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	82 self.params = [self.W, self.b]
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	83
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	84
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	85 class SigmoidalLayer(object):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	86 def __init__(self, rng, input, n_in, n_out):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	87 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	88 Typical hidden layer of a MLP: units are fully-connected and have
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	89 sigmoidal activation function. Weight matrix W is of shape (n_in,n_out)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	90 and the bias vector b is of shape (n_out,).
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	91
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	92 Hidden unit activation is given by: sigmoid(dot(input,W) + b)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	93
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	94 :type rng: numpy.random.RandomState
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	95 :param rng: a random number generator used to initialize weights
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	96 :type input: theano.tensor.dmatrix
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	97 :param input: a symbolic tensor of shape (n_examples, n_in)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	98 :type n_in: int
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	99 :param n_in: dimensionality of input
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	100 :type n_out: int
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	101 :param n_out: number of hidden units
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	102 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	103 self.input = input
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	104
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	105 W_values = numpy.asarray( rng.uniform( \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	106 low = -numpy.sqrt(6./(n_in+n_out)), \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	107 high = numpy.sqrt(6./(n_in+n_out)), \
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	108 size = (n_in, n_out)), dtype = theano.config.floatX)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	109 self.W = theano.shared(value = W_values)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	110
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	111 b_values = numpy.zeros((n_out,), dtype= theano.config.floatX)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	112 self.b = theano.shared(value= b_values)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	113
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	114 self.output = T.tanh(T.dot(input, self.W) + self.b)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	115 self.params = [self.W, self.b]
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	116
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	117
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	118 class LogisticRegression(object):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	119 """Multi-class Logistic Regression Class
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	120
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	121 The logistic regression is fully described by a weight matrix :math:`W`
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	122 and bias vector :math:`b`. Classification is done by projecting data
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	123 points onto a set of hyperplanes, the distance to which is used to
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	124 determine a class membership probability.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	125 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	126
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	127 def __init__(self, input, n_in, n_out):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	128 """ Initialize the parameters of the logistic regression
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	129 :param input: symbolic variable that describes the input of the
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	130 architecture (one minibatch)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	131 :type n_in: int
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	132 :param n_in: number of input units, the dimension of the space in
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	133 which the datapoints lie
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	134 :type n_out: int
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	135 :param n_out: number of output units, the dimension of the space in
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	136 which the labels lie
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	137 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	138
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	139 # initialize with 0 the weights W as a matrix of shape (n_in, n_out)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	140 self.W = theano.shared( value=numpy.zeros((n_in,n_out),
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	141 dtype = theano.config.floatX) )
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	142 # initialize the baises b as a vector of n_out 0s
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	143 self.b = theano.shared( value=numpy.zeros((n_out,),
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	144 dtype = theano.config.floatX) )
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	145 # compute vector of class-membership probabilities in symbolic form
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	146 self.p_y_given_x = T.nnet.softmax(T.dot(input, self.W)+self.b)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	147
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	148 # compute prediction as class whose probability is maximal in
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	149 # symbolic form
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	150 self.y_pred=T.argmax(self.p_y_given_x, axis=1)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	151
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	152 # list of parameters for this layer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	153 self.params = [self.W, self.b]
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	154
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	155 def negative_log_likelihood(self, y):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	156 """Return the mean of the negative log-likelihood of the prediction
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	157 of this model under a given target distribution.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	158 :param y: corresponds to a vector that gives for each example the
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	159 correct label
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	160 Note: we use the mean instead of the sum so that
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	161 the learning rate is less dependent on the batch size
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	162 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	163 return -T.mean(T.log(self.p_y_given_x)[T.arange(y.shape[0]),y])
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	164
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	165 def errors(self, y):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	166 """Return a float representing the number of errors in the minibatch
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	167 over the total number of examples of the minibatch ; zero one
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	168 loss over the size of the minibatch
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	169 """
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	170 # check if y has same dimension of y_pred
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	171 if y.ndim != self.y_pred.ndim:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	172 raise TypeError('y should have the same shape as self.y_pred',
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	173 ('y', target.type, 'y_pred', self.y_pred.type))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	174
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	175 # check if y is of the correct datatype
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	176 if y.dtype.startswith('int'):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	177 # the T.neq operator returns a vector of 0s and 1s, where 1
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	178 # represents a mistake in prediction
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	179 return T.mean(T.neq(self.y_pred, y))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	180 else:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	181 raise NotImplementedError()
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	182
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	183
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	184 def evaluate_lenet5(learning_rate=0.1, n_iter=200, batch_size=20, n_kern0=20, n_kern1=50, n_layer=3, filter_shape0=5, filter_shape1=5, sigmoide_size=500, dataset='mnist.pkl.gz'):
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	185 rng = numpy.random.RandomState(23455)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	186
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	187 print 'Before load dataset'
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	188 dataset=datasets.nist_digits
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	189 train_batches= dataset.train(batch_size)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	190 valid_batches=dataset.valid(batch_size)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	191 test_batches=dataset.test(batch_size)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	192 #print valid_batches.shape
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	193 #print test_batches.shape
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	194 print 'After load dataset'
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	195
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	196 ishape = (32,32) # this is the size of NIST images
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	197 n_kern2=80
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	198 n_kern3=100
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	199 if n_layer==4:
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	200 filter_shape1=3
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	201 filter_shape2=3
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	202 if n_layer==5:
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	203 filter_shape0=4
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	204 filter_shape1=2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	205 filter_shape2=2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	206 filter_shape3=2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	207
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	208
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	209 # allocate symbolic variables for the data
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	210 x = T.matrix('x') # rasterized images
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	211 y = T.lvector() # the labels are presented as 1D vector of [long int] labels
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	212
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	213
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	214 ######################
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	215 # BUILD ACTUAL MODEL #
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	216 ######################
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	217
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	218 # Reshape matrix of rasterized images of shape (batch_size,28*28)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	219 # to a 4D tensor, compatible with our LeNetConvPoolLayer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	220 layer0_input = x.reshape((batch_size,1,32,32))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	221
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	222 # Construct the first convolutional pooling layer:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	223 # filtering reduces the image size to (32-5+1,32-5+1)=(28,28)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	224 # maxpooling reduces this further to (28/2,28/2) = (14,14)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	225 # 4D output tensor is thus of shape (20,20,14,14)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	226 layer0 = LeNetConvPoolLayer(rng, input=layer0_input,
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	227 image_shape=(batch_size,1,32,32),
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	228 filter_shape=(n_kern0,1,filter_shape0,filter_shape0), poolsize=(2,2))
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	229
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	230 if(n_layer>2):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	231
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	232 # Construct the second convolutional pooling layer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	233 # filtering reduces the image size to (14-5+1,14-5+1)=(10,10)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	234 # maxpooling reduces this further to (10/2,10/2) = (5,5)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	235 # 4D output tensor is thus of shape (20,50,5,5)
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	236 fshape0=(32-filter_shape0+1)/2
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	237 layer1 = LeNetConvPoolLayer(rng, input=layer0.output,
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	238 image_shape=(batch_size,n_kern0,fshape0,fshape0),
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	239 filter_shape=(n_kern1,n_kern0,filter_shape1,filter_shape1), poolsize=(2,2))
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	240
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	241 else:
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	242
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	243 fshape0=(32-filter_shape0+1)/2
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	244 layer1_input = layer0.output.flatten(2)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	245 # construct a fully-connected sigmoidal layer
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	246 layer1 = SigmoidalLayer(rng, input=layer1_input,n_in=n_kern0fshape0fshape0, n_out=sigmoide_size)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	247
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	248 layer2 = LogisticRegression(input=layer1.output, n_in=sigmoide_size, n_out=10)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	249 cost = layer2.negative_log_likelihood(y)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	250 test_model = theano.function([x,y], layer2.errors(y))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	251 params = layer2.params+ layer1.params + layer0.params
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	252
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	253
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	254 if(n_layer>3):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	255
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	256 fshape0=(32-filter_shape0+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	257 fshape1=(fshape0-filter_shape1+1)/2
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	258 layer2 = LeNetConvPoolLayer(rng, input=layer1.output,
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	259 image_shape=(batch_size,n_kern1,fshape1,fshape1),
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	260 filter_shape=(n_kern2,n_kern1,filter_shape2,filter_shape2), poolsize=(2,2))
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	261
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	262 if(n_layer>4):
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	263
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	264
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	265 fshape0=(32-filter_shape0+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	266 fshape1=(fshape0-filter_shape1+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	267 fshape2=(fshape1-filter_shape2+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	268 fshape3=(fshape2-filter_shape3+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	269 layer3 = LeNetConvPoolLayer(rng, input=layer2.output,
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	270 image_shape=(batch_size,n_kern2,fshape2,fshape2),
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	271 filter_shape=(n_kern3,n_kern2,filter_shape3,filter_shape3), poolsize=(2,2))
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	272
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	273 layer4_input = layer3.output.flatten(2)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	274
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	275 layer4 = SigmoidalLayer(rng, input=layer4_input,
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	276 n_in=n_kern3fshape3fshape3, n_out=sigmoide_size)
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	277
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	278
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	279 layer5 = LogisticRegression(input=layer4.output, n_in=sigmoide_size, n_out=10)
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	280
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	281 cost = layer5.negative_log_likelihood(y)
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	282
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	283 test_model = theano.function([x,y], layer5.errors(y))
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	284
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	285 params = layer5.params+ layer4.params+ layer3.params+ layer2.params+ layer1.params + layer0.params
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	286
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	287 elif(n_layer>3):
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	288
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	289 fshape0=(32-filter_shape0+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	290 fshape1=(fshape0-filter_shape1+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	291 fshape2=(fshape1-filter_shape2+1)/2
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	292 layer3_input = layer2.output.flatten(2)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	293
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	294 layer3 = SigmoidalLayer(rng, input=layer3_input,
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	295 n_in=n_kern2fshape2fshape2, n_out=sigmoide_size)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	296
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	297
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	298 layer4 = LogisticRegression(input=layer3.output, n_in=sigmoide_size, n_out=10)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	299
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	300 cost = layer4.negative_log_likelihood(y)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	301
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	302 test_model = theano.function([x,y], layer4.errors(y))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	303
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	304 params = layer4.params+ layer3.params+ layer2.params+ layer1.params + layer0.params
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	305
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	306
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	307 elif(n_layer>2):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	308
253 a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	309 fshape0=(32-filter_shape0+1)/2
a491d3600a77 Derniere version du reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: 196 diff changeset	310 fshape1=(fshape0-filter_shape1+1)/2
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	311
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	312 # the SigmoidalLayer being fully-connected, it operates on 2D matrices of
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	313 # shape (batch_size,num_pixels) (i.e matrix of rasterized images).
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	314 # This will generate a matrix of shape (20,3244) = (20,512)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	315 layer2_input = layer1.output.flatten(2)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	316
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	317 # construct a fully-connected sigmoidal layer
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	318 layer2 = SigmoidalLayer(rng, input=layer2_input,
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	319 n_in=n_kern1fshape1fshape1, n_out=sigmoide_size)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	320
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	321
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	322 # classify the values of the fully-connected sigmoidal layer
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	323 layer3 = LogisticRegression(input=layer2.output, n_in=sigmoide_size, n_out=10)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	324
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	325 # the cost we minimize during training is the NLL of the model
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	326 cost = layer3.negative_log_likelihood(y)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	327
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	328 # create a function to compute the mistakes that are made by the model
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	329 test_model = theano.function([x,y], layer3.errors(y))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	330
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	331 # create a list of all model parameters to be fit by gradient descent
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	332 params = layer3.params+ layer2.params+ layer1.params + layer0.params
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	333
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	334
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	335
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	336
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	337
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	338 # create a list of gradients for all model parameters
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	339 grads = T.grad(cost, params)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	340
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	341 # train_model is a function that updates the model parameters by SGD
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	342 # Since this model has many parameters, it would be tedious to manually
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	343 # create an update rule for each model parameter. We thus create the updates
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	344 # dictionary by automatically looping over all (params[i],grads[i]) pairs.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	345 updates = {}
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	346 for param_i, grad_i in zip(params, grads):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	347 updates[param_i] = param_i - learning_rate * grad_i
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	348 train_model = theano.function([x, y], cost, updates=updates)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	349
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	350
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	351 ###############
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	352 # TRAIN MODEL #
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	353 ###############
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	354
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	355 #n_minibatches = len(train_batches)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	356 n_minibatches=0
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	357 n_valid=0
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	358 n_test=0
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	359 for x, y in dataset.train(batch_size):
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	360 if x.shape[0] == batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	361 n_minibatches+=1
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	362 n_minibatches*=batch_size
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	363 print n_minibatches
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	364
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	365 for x, y in dataset.valid(batch_size):
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	366 if x.shape[0] == batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	367 n_valid+=1
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	368 n_valid*=batch_size
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	369 print n_valid
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	370
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	371 for x, y in dataset.test(batch_size):
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	372 if x.shape[0] == batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	373 n_test+=1
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	374 n_test*=batch_size
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	375 print n_test
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	376
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	377
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	378 # early-stopping parameters
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	379 patience = 10000 # look as this many examples regardless
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	380 patience_increase = 2 # wait this much longer when a new best is
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	381 # found
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	382 improvement_threshold = 0.995 # a relative improvement of this much is
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	383 # considered significant
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	384 validation_frequency = n_minibatches # go through this many
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	385 # minibatche before checking the network
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	386 # on the validation set; in this case we
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	387 # check every epoch
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	388
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	389 best_params = None
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	390 best_validation_loss = float('inf')
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	391 best_iter = 0
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	392 test_score = 0.
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	393 start_time = time.clock()
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	394
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	395
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	396 # have a maximum of `n_iter` iterations through the entire dataset
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	397 iter=0
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	398 for epoch in xrange(n_iter):
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	399 for x, y in train_batches:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	400 if x.shape[0] != batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	401 continue
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	402 iter+=1
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	403
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	404 # get epoch and minibatch index
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	405 #epoch = iter / n_minibatches
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	406 minibatch_index = iter % n_minibatches
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	407
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	408 if iter %100 == 0:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	409 print 'training @ iter = ', iter
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	410 cost_ij = train_model(x,y)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	411
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	412
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	413 # compute zero-one loss on validation set
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	414 this_validation_loss = 0.
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	415 for x,y in valid_batches:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	416 if x.shape[0] != batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	417 continue
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	418 # sum up the errors for each minibatch
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	419 this_validation_loss += test_model(x,y)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	420
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	421 # get the average by dividing with the number of minibatches
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	422 this_validation_loss /= n_valid
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	423 print('epoch %i, minibatch %i/%i, validation error %f %%' % \
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	424 (epoch, minibatch_index+1, n_minibatches, \
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	425 this_validation_loss*100.))
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	426
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	427
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	428 # if we got the best validation score until now
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	429 if this_validation_loss < best_validation_loss:
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	430
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	431 #improve patience if loss improvement is good enough
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	432 if this_validation_loss < best_validation_loss * \
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	433 improvement_threshold :
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	434 patience = max(patience, iter * patience_increase)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	435
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	436 # save best validation score and iteration number
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	437 best_validation_loss = this_validation_loss
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	438 best_iter = iter
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	439
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	440 # test it on the test set
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	441 test_score = 0.
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	442 for x,y in test_batches:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	443 if x.shape[0] != batch_size:
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	444 continue
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	445 test_score += test_model(x,y)
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	446 test_score /= n_test
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	447 print((' epoch %i, minibatch %i/%i, test error of best '
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	448 'model %f %%') %
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	449 (epoch, minibatch_index+1, n_minibatches,
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	450 test_score*100.))
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	451
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	452 if patience <= iter :
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	453 break
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	454
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	455 end_time = time.clock()
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	456 print('Optimization complete.')
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	457 print('Best validation score of %f %% obtained at iteration %i,'\
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	458 'with test performance %f %%' %
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	459 (best_validation_loss * 100., best_iter, test_score*100.))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	460 print('The code ran for %f minutes' % ((end_time-start_time)/60.))
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	461
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	462 return (best_validation_loss * 100., test_score*100., (end_time-start_time)/60., best_iter)
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	463
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	464 if __name__ == '__main__':
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	465 evaluate_lenet5()
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	466
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	467 def experiment(state, channel):
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	468 print 'start experiment'
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	469 (best_validation_loss, test_score, minutes_trained, iter) = evaluate_lenet5(state.learning_rate, state.n_iter, state.batch_size, state.n_kern0, state.n_kern1, state.n_layer, state.filter_shape0, state.filter_shape1,state.sigmoide_size)
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	470 print 'end experiment'
270 d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	471
d41fe003fade Reseau a convolution avec le bon dataset Jeremy Eustache <jeremy.eustache@voila.fr> parents: 253 diff changeset	472 pylearn.version.record_versions(state,[theano,ift6266,pylearn])
146 33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	473
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	474 state.best_validation_loss = best_validation_loss
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	475 state.test_score = test_score
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	476 state.minutes_trained = minutes_trained
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	477 state.iter = iter
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	478
33038ab4e799 Reseau a convolution Jeremy Eustache <jeremy.eustache@voila.fr> parents: diff changeset	479 return channel.COMPLETE

Mercurial > ift6266

annotate baseline/conv_mlp/convolutional_mlp.py @ 352:cfb79f9fd1a4