Skip to content

Commit 31226dd

Browse files
nouiznouiz
committed
Merge pull request lisa-lab#50 from lisa-lab/fix_size_in_convnet_comment
Fix size in convnet comment
2 parents 3aa50bf + cee690d commit 31226dd

File tree

14 files changed

+104
-121
lines changed

14 files changed

+104
-121
lines changed

‎code/DBN.py

Lines changed: 3 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,5 @@
11
"""
22
"""
3-
import cPickle
4-
import gzip
53
import os
64
import sys
75
import time
@@ -372,7 +370,6 @@ def test_DBN(finetune_lr=0.1, pretraining_epochs=100,
372370
# on the validation set; in this case we
373371
# check every epoch
374372

375-
best_params = None
376373
best_validation_loss = numpy.inf
377374
test_score = 0.
378375
start_time = time.clock()
@@ -430,9 +427,10 @@ def test_DBN(finetune_lr=0.1, pretraining_epochs=100,
430427
end_time = time.clock()
431428
print(
432429
(
433-
'Optimization complete with best validation score of %f %%,'
430+
'Optimization complete with best validation score of %f %%, '
431+
'obtained at iteration %i, '
434432
'with test performance %f %%'
435-
) % (best_validation_loss * 100., test_score * 100.)
433+
) % (best_validation_loss * 100., best_iter + 1, test_score * 100.)
436434
)
437435
print >> sys.stderr, ('The fine tuning code for file ' +
438436
os.path.split(__file__)[1] +

‎code/SdA.py

Lines changed: 3 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -29,8 +29,6 @@
2929
Systems 19, 2007
3030
3131
"""
32-
import cPickle
33-
import gzip
3432
import os
3533
import sys
3634
import time
@@ -202,8 +200,6 @@ def pretraining_functions(self, train_set_x, batch_size):
202200
index = T.lscalar('index') # index to a minibatch
203201
corruption_level = T.scalar('corruption') # % of corruption to use
204202
learning_rate = T.scalar('lr') # learning rate to use
205-
# number of batches
206-
n_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
207203
# begining of a batch, given `index`
208204
batch_begin = index * batch_size
209205
# ending of a batch given `index`
@@ -429,7 +425,6 @@ def test_SdA(finetune_lr=0.1, pretraining_epochs=15,
429425
# on the validation set; in this case we
430426
# check every epoch
431427

432-
best_params = None
433428
best_validation_loss = numpy.inf
434429
test_score = 0.
435430
start_time = time.clock()
@@ -479,10 +474,11 @@ def test_SdA(finetune_lr=0.1, pretraining_epochs=15,
479474
end_time = time.clock()
480475
print(
481476
(
482-
'Optimization complete with best validation score of %f %%,'
477+
'Optimization complete with best validation score of %f %%, '
478+
'on iteration %i, '
483479
'with test performance %f %%'
484480
)
485-
% (best_validation_loss * 100., test_score * 100.)
481+
% (best_validation_loss * 100., best_iter + 1, test_score * 100.)
486482
)
487483
print >> sys.stderr, ('The training code for file ' +
488484
os.path.split(__file__)[1] +

‎code/cA.py

Lines changed: 8 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,8 @@
1212
squared Frobenius norm of the Jacobian of the hidden mapping h with
1313
respect to the visible units yields the contractive auto-encoder:
1414
15-
- \sum_{k=1}^d[ x_k \log z_k + (1-x_k) \log( 1-z_k)] + \| \frac{\partial h(x)}{\partial x} \|^2
15+
- \sum_{k=1}^d[ x_k \log z_k + (1-x_k) \log( 1-z_k)]
16+
+ \| \frac{\partial h(x)}{\partial x} \|^2
1617
1718
References :
1819
- S. Rifai, P. Vincent, X. Muller, X. Glorot, Y. Bengio: Contractive
@@ -27,8 +28,6 @@
2728
Systems 19, 2007
2829
2930
"""
30-
import cPickle
31-
import gzip
3231
import os
3332
import sys
3433
import time
@@ -79,11 +78,11 @@ class cA(object):
7978

8079
def __init__(self, numpy_rng, input=None, n_visible=784, n_hidden=100,
8180
n_batchsize=1, W=None, bhid=None, bvis=None):
82-
"""Initialize the cA class by specifying the number of visible units (the
83-
dimension d of the input ), the number of hidden units ( the dimension
84-
d' of the latent or hidden space ) and the contraction level. The
85-
constructor also receives symbolic variables for the input, weights and
86-
bias.
81+
"""Initialize the cA class by specifying the number of visible units
82+
(the dimension d of the input), the number of hidden units (the
83+
dimension d' of the latent or hidden space) and the contraction level.
84+
The constructor also receives symbolic variables for the input, weights
85+
and bias.
8786
8887
:type numpy_rng: numpy.random.RandomState
8988
:param numpy_rng: number random generator used to generate weights
@@ -161,7 +160,7 @@ def __init__(self, numpy_rng, input=None, n_visible=784, n_hidden=100,
161160
self.W_prime = self.W.T
162161

163162
# if no input is given, generate a variable representing the input
164-
if input == None:
163+
if input is None:
165164
# we use a matrix because we expect a minibatch of several
166165
# examples, each example being a row
167166
self.x = T.dmatrix(name='input')

‎code/convolutional_mlp.py

Lines changed: 16 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -21,8 +21,6 @@
2121
http://yann.lecun.com/exdb/publis/pdf/lecun-98.pdf
2222
2323
"""
24-
import cPickle
25-
import gzip
2624
import os
2725
import sys
2826
import time
@@ -53,14 +51,14 @@ def __init__(self, rng, input, filter_shape, image_shape, poolsize=(2, 2)):
5351
5452
:type filter_shape: tuple or list of length 4
5553
:param filter_shape: (number of filters, num input feature maps,
56-
filter height,filter width)
54+
filter height, filter width)
5755
5856
:type image_shape: tuple or list of length 4
5957
:param image_shape: (batch size, num input feature maps,
6058
image height, image width)
6159
6260
:type poolsize: tuple or list of length 2
63-
:param poolsize: the downsampling (pooling) factor (#rows,#cols)
61+
:param poolsize: the downsampling (pooling) factor (#rows, #cols)
6462
"""
6563

6664
assert image_shape[1] == filter_shape[1]
@@ -104,7 +102,7 @@ def __init__(self, rng, input, filter_shape, image_shape, poolsize=(2, 2)):
104102
)
105103

106104
# add the bias term. Since the bias is a vector (1D array), we first
107-
# reshape it to a tensor of shape (1,n_filters,1,1). Each bias will
105+
# reshape it to a tensor of shape (1, n_filters, 1, 1). Each bias will
108106
# thus be broadcasted across mini-batches and feature map
109107
# width & height
110108
self.output = T.tanh(pooled_out + self.b.dimshuffle('x', 0, 'x', 'x'))
@@ -155,21 +153,21 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
155153
x = T.matrix('x') # the data is presented as rasterized images
156154
y = T.ivector('y') # the labels are presented as 1D vector of
157155
# [int] labels
158-
ishape = (28, 28) # this is the size of MNIST images
159156

160157
######################
161158
# BUILD ACTUAL MODEL #
162159
######################
163160
print '... building the model'
164161

165-
# Reshape matrix of rasterized images of shape (batch_size,28*28)
162+
# Reshape matrix of rasterized images of shape (batch_size, 28 * 28)
166163
# to a 4D tensor, compatible with our LeNetConvPoolLayer
164+
# (28, 28) is the size of MNIST images.
167165
layer0_input = x.reshape((batch_size, 1, 28, 28))
168166

169167
# Construct the first convolutional pooling layer:
170-
# filtering reduces the image size to (28-5+1,28-5+1)=(24,24)
171-
# maxpooling reduces this further to (24/2,24/2) = (12,12)
172-
# 4D output tensor is thus of shape (batch_size,nkerns[0],12,12)
168+
# filtering reduces the image size to (28-5+1 , 28-5+1) = (24, 24)
169+
# maxpooling reduces this further to (24/2, 24/2) = (12, 12)
170+
# 4D output tensor is thus of shape (batch_size, nkerns[0], 12, 12)
173171
layer0 = LeNetConvPoolLayer(
174172
rng,
175173
input=layer0_input,
@@ -179,9 +177,9 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
179177
)
180178

181179
# Construct the second convolutional pooling layer
182-
# filtering reduces the image size to (12-5+1,12-5+1)=(8,8)
183-
# maxpooling reduces this further to (8/2,8/2) = (4,4)
184-
# 4D output tensor is thus of shape (nkerns[0],nkerns[1],4,4)
180+
# filtering reduces the image size to (12-5+1, 12-5+1) = (8, 8)
181+
# maxpooling reduces this further to (8/2, 8/2) = (4, 4)
182+
# 4D output tensor is thus of shape (nkerns[0], nkerns[1], 4, 4)
185183
layer1 = LeNetConvPoolLayer(
186184
rng,
187185
input=layer0.output,
@@ -191,8 +189,9 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
191189
)
192190

193191
# the HiddenLayer being fully-connected, it operates on 2D matrices of
194-
# shape (batch_size,num_pixels) (i.e matrix of rasterized images).
195-
# This will generate a matrix of shape (20,32*4*4) = (20,512)
192+
# shape (batch_size, num_pixels) (i.e matrix of rasterized images).
193+
# This will generate a matrix of shape (batch_size, nkerns[1] * 4 * 4),
194+
# or (500, 50 * 4 * 4) = (500, 800) with the default values.
196195
layer2_input = layer1.output.flatten(2)
197196

198197
# construct a fully-connected sigmoidal layer
@@ -239,7 +238,7 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
239238
# SGD Since this model has many parameters, it would be tedious to
240239
# manually create an update rule for each model parameter. We thus
241240
# create the updates list by automatically looping over all
242-
# (params[i],grads[i]) pairs.
241+
# (params[i], grads[i]) pairs.
243242
updates = [
244243
(param_i, param_i - learning_rate * grad_i)
245244
for param_i, grad_i in zip(params, grads)
@@ -272,7 +271,6 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
272271
# on the validation set; in this case we
273272
# check every epoch
274273

275-
best_params = None
276274
best_validation_loss = numpy.inf
277275
best_iter = 0
278276
test_score = 0.
@@ -330,7 +328,7 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
330328

331329
end_time = time.clock()
332330
print('Optimization complete.')
333-
print('Best validation score of %f %% obtained at iteration %i,'
331+
print('Best validation score of %f %% obtained at iteration %i, '
334332
'with test performance %f %%' %
335333
(best_validation_loss * 100., best_iter + 1, test_score * 100.))
336334
print >> sys.stderr, ('The code for file ' +

‎code/dA.py

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -30,8 +30,6 @@
3030
3131
"""
3232

33-
import cPickle
34-
import gzip
3533
import os
3634
import sys
3735
import time
@@ -185,7 +183,7 @@ def __init__(
185183
self.W_prime = self.W.T
186184
self.theano_rng = theano_rng
187185
# if no input is given, generate a variable representing the input
188-
if input == None:
186+
if input is None:
189187
# we use a matrix because we expect a minibatch of several
190188
# examples, each example being a row
191189
self.x = T.dmatrix(name='input')

0 commit comments

Comments
 (0)