view scripts/stacked_dae.py @ 116:3bec123dd75d

changes on pipeline mecanism: we now sample a different complexity for each transformations, this because when we use the same sampled complexity for all the modules 1/8 of the time we are close to 0 and we obtain an image very close to the source, we now save a complexity for each module in the parameters array
author Xavier Glorot <glorotxa@iro.umontreal.ca>
date Wed, 17 Feb 2010 16:22:54 -0500
parents 0b4080394f2c
children 4f37755d301b
line wrap: on
line source

#!/usr/bin/python
# coding: utf-8

# Code for stacked denoising autoencoder
# Tests with MNIST
# TODO: adapt for NIST
# Based almost entirely on deeplearning.net tutorial, modifications by
# François Savard

# Base LogisticRegression, SigmoidalLayer, dA, SdA code taken
# from the deeplearning.net tutorial. Refactored a bit.
# Changes (mainly):
# - splitted initialization in smaller methods
# - removed the "givens" thing involving an index in the whole dataset
#       (to allow flexibility in how data is inputted... not necessarily one big tensor)
# - changed the "driver" a lot, altough for the moment the same logic is used

import time
import theano
import theano.tensor as T
import theano.tensor.nnet
from theano.tensor.shared_randomstreams import RandomStreams
import numpy, numpy.random

from pylearn.datasets import MNIST


# from pylearn codebase
def update_locals(obj, dct):
    if 'self' in dct:
        del dct['self']
    obj.__dict__.update(dct)


class LogisticRegression(object):
    def __init__(self, input, n_in, n_out):
        # initialize with 0 the weights W as a matrix of shape (n_in, n_out) 
        self.W = theano.shared(value=numpy.zeros((n_in,n_out), dtype = theano.config.floatX),
                                name='W')
        # initialize the baises b as a vector of n_out 0s
        self.b = theano.shared(value=numpy.zeros((n_out,), dtype = theano.config.floatX),
                               name='b')

        # compute vector of class-membership probabilities in symbolic form
        self.p_y_given_x = T.nnet.softmax(T.dot(input, self.W)+self.b)

        # compute prediction as class whose probability is maximal in 
        # symbolic form
        self.y_pred=T.argmax(self.p_y_given_x, axis=1)

        self.params = [self.W, self.b]

    def negative_log_likelihood(self, y):
        return -T.mean(T.log(self.p_y_given_x)[T.arange(y.shape[0]),y])

    def errors(self, y):
        # check if y has same dimension of y_pred 
        if y.ndim != self.y_pred.ndim:
            raise TypeError('y should have the same shape as self.y_pred', 
                ('y', target.type, 'y_pred', self.y_pred.type))
        # check if y is of the correct datatype        
        if y.dtype.startswith('int'):
            # the T.neq operator returns a vector of 0s and 1s, where 1
            # represents a mistake in prediction
            return T.mean(T.neq(self.y_pred, y))
        else:
            raise NotImplementedError()


class SigmoidalLayer(object):
    def __init__(self, rng, input, n_in, n_out):
        self.input = input

        W_values = numpy.asarray( rng.uniform( \
              low = -numpy.sqrt(6./(n_in+n_out)), \
              high = numpy.sqrt(6./(n_in+n_out)), \
              size = (n_in, n_out)), dtype = theano.config.floatX)
        self.W = theano.shared(value = W_values)

        b_values = numpy.zeros((n_out,), dtype= theano.config.floatX)
        self.b = theano.shared(value= b_values)

        self.output = T.nnet.sigmoid(T.dot(input, self.W) + self.b)
        self.params = [self.W, self.b]


class dA(object):
    def __init__(self, n_visible= 784, n_hidden= 500, \
              corruption_level = 0.1, input = None, \
              shared_W = None, shared_b = None):
        update_locals(self, locals())

        self.init_randomizer()
        self.init_params()
        self.init_functions()

    def init_randomizer(self):
        # create a Theano random generator that gives symbolic random values
        self.theano_rng = RandomStreams()
        # create a numpy random generator
        self.numpy_rng = numpy.random.RandomState()

    def init_params(self):
        if self.shared_W != None and self.shared_b != None :
            self.W = self.shared_W
            self.b = self.shared_b
        else:
            # initial values for weights and biases
            # note : W' was written as `W_prime` and b' as `b_prime`

            # W is initialized with `initial_W` which is uniformely sampled
            # from -6./sqrt(n_visible+n_hidden) and 6./sqrt(n_hidden+n_visible)
            # the output of uniform if converted using asarray to dtype
            # theano.config.floatX so that the code is runable on GPU
            initial_W = numpy.asarray( self.numpy_rng.uniform( \
                 low  = -numpy.sqrt(6./(n_hidden+n_visible)), \
                 high = numpy.sqrt(6./(n_hidden+n_visible)), \
                 size = (n_visible, n_hidden)), dtype = theano.config.floatX)
            initial_b = numpy.zeros(n_hidden)

            # theano shared variables for weights and biases
            self.W = theano.shared(value = initial_W, name = "W")
            self.b = theano.shared(value = initial_b, name = "b")

        initial_b_prime= numpy.zeros(self.n_visible)
        # tied weights, therefore W_prime is W transpose
        self.W_prime = self.W.T
        self.b_prime = theano.shared(value = initial_b_prime, name = "b'")

    def init_functions(self):
        # if no input is given, generate a variable representing the input
        if self.input == None :
            # we use a matrix because we expect a minibatch of several examples,
            # each example being a row
            self.x = T.dmatrix(name = 'input')
        else:
            self.x = self.input

        # keep 90% of the inputs the same and zero-out randomly selected subset of 
        # 10% of the inputs
        # note : first argument of theano.rng.binomial is the shape(size) of
        #        random numbers that it should produce
        #        second argument is the number of trials
        #        third argument is the probability of success of any trial
        #
        #        this will produce an array of 0s and 1s where 1 has a
        #        probability of 1 - ``corruption_level`` and 0 with
        #        ``corruption_level``
        self.tilde_x = self.theano_rng.binomial(self.x.shape, 1, 1-self.corruption_level) * self.x
        # using tied weights
        self.y = T.nnet.sigmoid(T.dot(self.tilde_x, self.W) + self.b)
        self.z = T.nnet.sigmoid(T.dot(self.y, self.W_prime) + self.b_prime)
        self.L = - T.sum( self.x*T.log(self.z) + (1-self.x)*T.log(1-self.z), axis=1 )
        # note : L is now a vector, where each element is the cross-entropy cost
        #        of the reconstruction of the corresponding example of the
        #        minibatch. We need to compute the average of all these to get
        #        the cost of the minibatch
        self.cost = T.mean(self.L)

        self.params = [ self.W, self.b, self.b_prime ]

class SdA():
    def __init__(self, batch_size, n_ins,
               hidden_layers_sizes, n_outs,
               corruption_levels, rng, pretrain_lr, finetune_lr):
        update_locals(self, locals())

        self.layers             = []
        self.pretrain_functions = []
        self.params             = []
        self.n_layers           = len(hidden_layers_sizes)

        if len(hidden_layers_sizes) < 1 :
            raiseException (' You must have at least one hidden layer ')

        # allocate symbolic variables for the data
        self.x = T.matrix('x')  # the data is presented as rasterized images
        self.y = T.ivector('y') # the labels are presented as 1D vector of
                               # [int] labels

        self.create_layers()
        self.init_finetuning()

    def create_layers(self):
        for i in xrange( self.n_layers ):
            # construct the sigmoidal layer

            # the size of the input is either the number of hidden units of
            # the layer below or the input size if we are on the first layer
            if i == 0 :
                input_size = self.n_ins
            else:
                input_size = self.hidden_layers_sizes[i-1]

            # the input to this layer is either the activation of the hidden
            # layer below or the input of the SdA if you are on the first
            # layer
            if i == 0 :
                layer_input = self.x
            else:
                layer_input = self.layers[-1].output

            layer = SigmoidalLayer(self.rng, layer_input, input_size,
                                   self.hidden_layers_sizes[i] )
            # add the layer to the
            self.layers += [layer]
            self.params += layer.params

            # Construct a denoising autoencoder that shared weights with this
            # layer
            dA_layer = dA(input_size, self.hidden_layers_sizes[i], \
                          corruption_level = self.corruption_levels[0],\
                          input = layer_input, \
                          shared_W = layer.W, shared_b = layer.b)

            self.init_updates_for_layer(dA_layer)

    def init_updates_for_layer(self, dA_layer):
        # Construct a function that trains this dA
        # compute gradients of layer parameters
        gparams = T.grad(dA_layer.cost, dA_layer.params)
        # compute the list of updates
        updates = {}
        for param, gparam in zip(dA_layer.params, gparams):
            updates[param] = param - gparam * self.pretrain_lr

        # create a function that trains the dA
        update_fn = theano.function([self.x], dA_layer.cost, \
              updates = updates)

        # collect this function into a list
        self.pretrain_functions += [update_fn]

    def init_finetuning(self):
        # We now need to add a logistic layer on top of the MLP
        self.logLayer = LogisticRegression(\
                         input = self.layers[-1].output,\
                         n_in = self.hidden_layers_sizes[-1], n_out = self.n_outs)

        self.params += self.logLayer.params
        # construct a function that implements one step of finetunining

        # compute the cost, defined as the negative log likelihood
        cost = self.logLayer.negative_log_likelihood(self.y)
        # compute the gradients with respect to the model parameters
        gparams = T.grad(cost, self.params)
        # compute list of updates
        updates = {}
        for param,gparam in zip(self.params, gparams):
            updates[param] = param - gparam*self.finetune_lr

        self.finetune = theano.function([self.x, self.y], cost,
                updates = updates)

        # symbolic variable that points to the number of errors made on the
        # minibatch given by self.x and self.y

        self.errors = self.logLayer.errors(self.y)

class MnistIterators:
    def __init__(self, minibatch_size):
        self.minibatch_size = minibatch_size

        self.mnist = MNIST.first_1k()

        self.len_train = len(self.mnist.train.x)
        self.len_valid = len(self.mnist.valid.x)
        self.len_test = len(self.mnist.test.x)

    def train_x_batches(self):
        idx = 0
        while idx < len(self.mnist.train.x):
            yield self.mnist.train.x[idx:idx+self.minibatch_size]
            idx += self.minibatch_size

    def train_xy_batches(self):
        idx = 0
        while idx < len(self.mnist.train.x):
            mb_x = self.mnist.train.x[idx:idx+self.minibatch_size]
            mb_y = self.mnist.train.y[idx:idx+self.minibatch_size]
            yield mb_x, mb_y
            idx += self.minibatch_size

    def valid_xy_batches(self):
        idx = 0
        while idx < len(self.mnist.valid.x):
            mb_x = self.mnist.valid.x[idx:idx+self.minibatch_size]
            mb_y = self.mnist.valid.y[idx:idx+self.minibatch_size]
            yield mb_x, mb_y
            idx += self.minibatch_size


class MnistTrainingDriver:
    def __init__(self, rng=numpy.random):
        self.rng = rng

        self.init_SdA()

    def init_SdA(self):
        # Hyperparam
        hidden_layers_sizes = [1000, 1000, 1000]
        n_outs = 10
        corruption_levels = [0.2, 0.2, 0.2]
        minibatch_size = 10
        pretrain_lr = 0.001
        finetune_lr = 0.001

        update_locals(self, locals())

        self.mnist = MnistIterators(minibatch_size)

        # construct the stacked denoising autoencoder class
        self.classifier = SdA( batch_size = minibatch_size, \
                          n_ins=28*28, \
                          hidden_layers_sizes = hidden_layers_sizes, \
                          n_outs=n_outs, \
                          corruption_levels = corruption_levels,\
                          rng = self.rng,\
                          pretrain_lr = pretrain_lr, \
                          finetune_lr = finetune_lr) 

    def compute_validation_error(self):
        validation_error = 0.0

        count = 0
        for mb_x, mb_y in self.mnist.valid_xy_batches():
            validation_error += self.classifier.errors(mb_x, mb_y)
            count += 1

        return float(validation_error) / count

    def pretrain(self):
        pretraining_epochs = 20

        for layer_idx, update_fn in enumerate(self.classifier.pretrain_functions):
            for epoch in xrange(pretraining_epochs):
                # go through the training set
                cost_acc = 0.0
                for i, mb_x in enumerate(self.mnist.train_x_batches()):
                    cost_acc += update_fn(mb_x)
                    
                    if i % 100 == 0:
                        print i, "avg err = ", cost_acc / 100.0
                        cost_acc = 0.0
                print 'Pre-training layer %d, epoch %d' % (layer_idx, epoch)

    def finetune(self):
        max_training_epochs = 1000

        n_train_batches = self.mnist.len_train / self.minibatch_size

        # early-stopping parameters
        patience = 10000 # look as this many examples regardless
        patience_increase = 2. # wait this much longer when a new best is
                                      # found
        improvement_threshold = 0.995 # a relative improvement of this much is
                                      # considered significant
        validation_frequency = min(n_train_batches, patience/2)
                                      # go through this many
                                      # minibatche before checking the network
                                      # on the validation set; in this case we
                                      # check every epoch
     
     
        # TODO: use this
        best_params = None
        best_validation_loss = float('inf')
        test_score = 0.
        start_time = time.clock()
     
        done_looping = False
        epoch = 0
     
        while (epoch < max_training_epochs) and (not done_looping):
            epoch = epoch + 1
            for minibatch_index, (mb_x, mb_y) in enumerate(self.mnist.train_xy_batches()):
                cost_ij = classifier.finetune(mb_x, mb_y)
                iter = epoch * n_train_batches + minibatch_index
         
                if (iter+1) % validation_frequency == 0:
                    this_validation_loss = self.compute_validation_error()
                    print('epoch %i, minibatch %i/%i, validation error %f %%' % \
                           (epoch, minibatch_index+1, n_train_batches, \
                            this_validation_loss*100.))
         
                    # if we got the best validation score until now
                    if this_validation_loss < best_validation_loss:
         
                        #improve patience if loss improvement is good enough
                        if this_validation_loss < best_validation_loss * \
                               improvement_threshold :
                            patience = max(patience, iter * patience_increase)
                            print "Improving patience"
         
                        # save best validation score and iteration number
                        best_validation_loss = this_validation_loss
                        best_iter = iter
         
                        # test it on the test set
                        #test_losses = [test_model(i) for i in xrange(n_test_batches)]
                        #test_score = numpy.mean(test_losses)
                        #print((' epoch %i, minibatch %i/%i, test error of best '
                        #      'model %f %%') %
                        #             (epoch, minibatch_index+1, n_train_batches,
                        #              test_score*100.))
     
     
            if patience <= iter :
                done_looping = True
                break

def train():
    driver = MnistTrainingDriver()
    start_time = time.clock()
    driver.pretrain()
    print "PRETRAINING DONE. STARTING FINETUNING."
    driver.finetune()
    end_time = time.clock()

if __name__ == '__main__':
    train()