ift6266: deep/stacked_dae/stacked

comparison deep/stacked_dae/stacked_dae.py @ 167:1f5937e9e530

More moves - transformations into data_generation, added "deep" folder

author	Dumitru Erhan <dumitru.erhan@gmail.com>
date	Fri, 26 Feb 2010 14:15:38 -0500
parents	scripts/stacked_dae/stacked_dae.py@7d8366fb90bf
children	b9ea8e2d071a

comparison

equal deleted inserted replaced

-:17ae5a1a4dd1
+:1f5937e9e530
+#!/usr/bin/python
+# coding: utf-8
+import numpy
+import theano
+import time
+import theano.tensor as T
+from theano.tensor.shared_randomstreams import RandomStreams
+import copy
+from utils import update_locals
+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) )
+# initialize the baises b as a vector of n_out 0s
+self.b = theano.shared( value=numpy.zeros((n_out,),
+dtype = theano.config.floatX) )
+# 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)
+# list of parameters for this layer
+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):
+self.n_visible = n_visible
+self.n_hidden  = n_hidden
+# create a Theano random generator that gives symbolic random values
+theano_rng = RandomStreams()
+if shared_W != None and shared_b != None :
+self.W = shared_W
+self.b = 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( numpy.random.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, dtype = theano.config.floatX)
+# 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(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'")
+# if no input is given, generate a variable representing the input
+if 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 = input
+# Equation (1)
+# 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  = theano_rng.binomial( self.x.shape,  1,  1 - corruption_level) * self.x
+# Equation (2)
+# note  : y is stored as an attribute of the class so that it can be
+#         used later when stacking dAs.
+self.y   = T.nnet.sigmoid(T.dot(self.tilde_x, self.W      ) + self.b)
+# Equation (3)
+self.z   = T.nnet.sigmoid(T.dot(self.y, self.W_prime) + self.b_prime)
+# Equation (4)
+# note : we sum over the size of a datapoint; if we are using minibatches,
+#        L will  be a vector, with one entry per example in minibatch
+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(object):
+def __init__(self, train_set_x, train_set_y, batch_size, n_ins,
+hidden_layers_sizes, n_outs,
+corruption_levels, rng, pretrain_lr, finetune_lr, input_divider=1.0):
+update_locals(self, locals())
+self.layers             = []
+self.pretrain_functions = []
+self.params             = []
+self.n_layers           = len(hidden_layers_sizes)
+self.input_divider = numpy.asarray(input_divider, dtype=theano.config.floatX)
+if len(hidden_layers_sizes) < 1 :
+raiseException (' You must have at least one hidden layer ')
+# allocate symbolic variables for the data
+index   = T.lscalar()    # index to a [mini]batch
+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
+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 = n_ins
+else:
+input_size = 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(rng, layer_input, input_size,
+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, hidden_layers_sizes[i], \
+corruption_level = corruption_levels[0],\
+input = layer_input, \
+shared_W = layer.W, shared_b = layer.b)
+# 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 * pretrain_lr
+# create a function that trains the dA
+update_fn = theano.function([index], dA_layer.cost, \
+updates = updates,
+givens = {
+self.x : train_set_x[index*batch_size:(index+1)*batch_size] / self.input_divider})
+# collect this function into a list
+self.pretrain_functions += [update_fn]
+# We now need to add a logistic layer on top of the MLP
+self.logLayer = LogisticRegression(\
+input = self.layers[-1].output,\
+n_in = hidden_layers_sizes[-1], n_out = 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*finetune_lr
+self.finetune = theano.function([index], cost,
+updates = updates,
+givens = {
+self.x : train_set_x[index*batch_size:(index+1)*batch_size]/self.input_divider,
+self.y : train_set_y[index*batch_size:(index+1)*batch_size]} )
+# 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)
+@classmethod
+def copy_reusing_lower_layers(cls, obj, num_hidden_layers, new_finetuning_lr=None):
+assert(num_hidden_layers <= obj.n_layers)
+if not new_finetuning_lr:
+new_finetuning_lr = obj.finetune_lr
+new_sda = cls(train_set_x= obj.train_set_x, \
+train_set_y = obj.train_set_y,\
+batch_size = obj.batch_size, \
+n_ins= obj.n_ins, \
+hidden_layers_sizes = obj.hidden_layers_sizes[:num_hidden_layers], \
+n_outs = obj.n_outs, \
+corruption_levels = obj.corruption_levels[:num_hidden_layers],\
+rng = obj.rng,\
+pretrain_lr = obj.pretrain_lr, \
+finetune_lr = new_finetuning_lr, \
+input_divider = obj.input_divider )
+# new_sda.layers contains only the hidden layers actually
+for i, layer in enumerate(new_sda.layers):
+original_layer = obj.layers[i]
+for p1,p2 in zip(layer.params, original_layer.params):
+p1.value = p2.value.copy()
+return new_sda
+def get_params_copy(self):
+return copy.deepcopy(self.params)
+def set_params_from_copy(self, copy):
+# We don't want to replace the var, as the functions have pointers in there
+# We only want to replace values.
+for i, p in enumerate(self.params):
+p.value = copy[i].value
+def get_params_means(self):
+s = []
+for p in self.params:
+s.append(numpy.mean(p.value))
+return s
+if __name__ == '__main__':
+import sys
+args = sys.argv[1:]

Mercurial > ift6266

comparison deep/stacked_dae/stacked_dae.py @ 167:1f5937e9e530