Skip to content

Commit

Permalink
Matching Goodfellow's GAN
Browse files Browse the repository at this point in the history
  • Loading branch information
Ragav Venkatesan committed Mar 5, 2017
1 parent 0d1f763 commit 4587d9c
Show file tree
Hide file tree
Showing 2 changed files with 50 additions and 96 deletions.
144 changes: 49 additions & 95 deletions pantry/tutorials/gan.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -268,15 +268,18 @@ def deep_gan (dataset, verbose = 1 ):
net: A Network object.
Notes:
This is not setup properly therefore does not learn at the moment.
This is not setup properly therefore does not learn at the moment. This network here mimics
Ian Goodfellow's original code and implementation for MNIST adapted from his source code:
https://github.com/goodfeli/adversarial/blob/master/mnist.yaml .It might not be a perfect
replicaiton, but I tried as best as I could.
"""
if verbose >=2:
print (".. Creating a GAN network")

optimizer_params = {
"momentum_type" : 'polyak',
"momentum_params" : (0.65, 0.9, 50),
"regularization" : (0.0001, 0.0001),
"momentum_params" : (0.5, 0.7, 20),
"regularization" : (0.000, 0.000),
"optimizer_type" : 'rmsprop',
"id" : "main"
}
Expand Down Expand Up @@ -314,10 +317,11 @@ def deep_gan (dataset, verbose = 1 ):
#z - latent space created by random layer
net.add_layer(type = 'random',
id = 'z',
num_neurons = (100,32),
num_neurons = (100,10),
distribution = 'normal',
mu = 0,
sigma = 1,
# limits = (0,1),
verbose = verbose)

#x - inputs come from dataset 1 X 784
Expand All @@ -332,7 +336,7 @@ def deep_gan (dataset, verbose = 1 ):
net.add_layer ( type = "dot_product",
origin = "z",
id = "G1",
num_neurons = 128,
num_neurons = 1200,
activation = 'relu',
# batch_norm = True,
verbose = verbose
Expand All @@ -341,7 +345,7 @@ def deep_gan (dataset, verbose = 1 ):
net.add_layer ( type = "dot_product",
origin = "G1",
id = "G2",
num_neurons = 384,
num_neurons = 1200,
activation = 'relu',
# batch_norm = True,
verbose = verbose
Expand All @@ -363,106 +367,57 @@ def deep_gan (dataset, verbose = 1 ):
shape = (28,28),
verbose = verbose )

net.add_layer ( type = "conv_pool",
id = "D1-x",
origin = "x",
num_neurons = 40,
filter_shape = (5,5),
pool_size = (2,2),
activation = ('maxout', 'maxout', 2),
regularize = True,
# batch_norm = True,
verbose = verbose
)

net.add_layer ( type = "conv_pool",
id = "D1-z",
origin = "G(z)-unflattened",
num_neurons = 40,
filter_shape = (5,5),
pool_size = (2,2),
activation = ('maxout', 'maxout', 2),
regularize = True,
# batch_norm = True,
input_params = net.dropout_layers['D1-x'].params,
verbose = verbose
)

net.add_layer ( type = "conv_pool",
id = "D2-x",
origin = "D1-x",
num_neurons = 100,
net = (3,3),
pool_size = (2,2),
activation = ('maxout', 'maxout', 2),
regularize = True,
# batch_norm = True,
verbose = verbose
)

net.add_layer ( type = "conv_pool",
id = "D2-z",
origin = "D1-z",
num_neurons = 100,
activation = ('maxout', 'maxout', 2),
filter_shape = (3,3),
pool_size = (2,2),
regularize = True,
# batch_norm = True,
input_params = net.dropout_layers['D2-x'].params,
verbose = verbose
)

net.add_layer ( type = "dot_product",
id = "D3-x",
origin = "D2-x",
num_neurons = 800,
activation = 'relu',
id = "D1-x",
origin = "x",
num_neurons = 1200,
activation = ('maxout','maxout',5),
regularize = True,
# batch_norm = True,
dropout_rate = 0.5,
# dropout_rate = 0.5,
verbose = verbose
)

net.add_layer ( type = "dot_product",
id = "D3-z",
origin = "D2-z",
input_params = net.dropout_layers["D3-x"].params,
num_neurons = 800,
activation = 'relu',
id = "D1-z",
origin = "G(z)-unflattened",
input_params = net.dropout_layers["D1-x"].params,
num_neurons = 1200,
activation = ('maxout','maxout',5),
regularize = True,
# batch_norm = True,
dropout_rate = 0.5,
# dropout_rate = 0.5,
verbose = verbose
)

net.add_layer ( type = "dot_product",
id = "D4-x",
origin = "D3-x",
num_neurons = 800,
activation = 'relu',
id = "D2-x",
origin = "D1-x",
num_neurons = 1200,
activation = ('maxout','maxout',5),
regularize = True,
# batch_norm = True,
dropout_rate = 0.5,
# dropout_rate = 0.5,
verbose = verbose
)

net.add_layer ( type = "dot_product",
id = "D4-z",
origin = "D3-z",
input_params = net.dropout_layers["D4-x"].params,
num_neurons = 800,
activation = 'relu',
# batch_norm = True,
id = "D2-z",
origin = "D1-z",
input_params = net.dropout_layers["D2-x"].params,
num_neurons = 1200,
activation = ('maxout','maxout',5),
regularize = True,
dropout_rate = 0.5,
# dropout_rate = 0.5,
# batch_norm = True,
verbose = verbose
)

#C(D(x)) - This is the opposite of C(D(G(z))), real
net.add_layer ( type = "dot_product",
id = "D(x)",
origin = "D4-x",
origin = "D2-x",
num_neurons = 1,
activation = 'sigmoid',
verbose = verbose
Expand All @@ -471,23 +426,23 @@ def deep_gan (dataset, verbose = 1 ):
#C(D(G(z))) fake - the classifier for fake/real that always predicts fake
net.add_layer ( type = "dot_product",
id = "D(G(z))",
origin = "D4-z",
origin = "D2-z",
num_neurons = 1,
activation = 'sigmoid',
input_params = net.dropout_layers["D(x)"].params,
verbose = verbose
)


"""
#C(D(x)) - This is the opposite of C(D(G(z))), real
net.add_layer ( type = "classifier",
id = "softmax",
origin = "D4-x",
origin = "D2-x",
num_classes = 10,
activation = 'softmax',
verbose = verbose
)

"""
# objective layers
# discriminator objective
net.add_layer (type = "tensor",
Expand Down Expand Up @@ -519,7 +474,7 @@ def deep_gan (dataset, verbose = 1 ):
datastream_origin = 'data',
verbose = verbose
)

"""
#softmax objective.
net.add_layer ( type = "objective",
id = "classifier_obj",
Expand All @@ -529,26 +484,25 @@ def deep_gan (dataset, verbose = 1 ):
datastream_origin = 'data',
verbose = verbose
)

from yann.utils.graph import draw_network
draw_network(net.graph, filename = 'gan.png')
"""
# from yann.utils.graph import draw_network
# draw_network(net.graph, filename = 'gan.png')
net.pretty_print()

net.cook ( objective_layers = ["classifier_obj", "discriminator_obj", \
"generator_obj"],
net.cook ( objective_layers = [None, "discriminator_obj", "generator_obj"],
optimizer_params = optimizer_params,
discriminator_layers = ["D1-x","D2-x","D3-x","D4-x"],
discriminator_layers = ["D1-x","D2-x"],
generator_layers = ["G1","G2","G(z)"],
classifier_layers = ["D1-x","D2-x","D3-x","D4-x", "softmax"],
softmax_layer = "softmax",
# classifier_layers = ["D1-x","D2-x","softmax"],
# softmax_layer = "softmax",
game_layers = ("D(x)", "D(G(z))"),
verbose = verbose )

learning_rates = (0.05, 0.01 )
learning_rates = (0.00004, 0.001 )

net.train( epochs = (20),
k = 5,
pre_train_discriminator = 3,
k = 1,
pre_train_discriminator = 0,
validate_after_epochs = 1,
visualize_after_epochs = 1,
training_accuracy = True,
Expand Down
2 changes: 1 addition & 1 deletion yann/special/gan.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -441,7 +441,7 @@ def _create_layer_activities(self, datastream = None, verbose = 2):

index = T.lscalar('index')
self.layer_activities_created = True
for id, _layer in self.layers.iteritems():
for id, _layer in self.inference_layers.iteritems():
if verbose >=3 :
print("... collecting the activities of layer " + id)
activity = _layer.output
Expand Down

0 comments on commit 4587d9c

Please sign in to comment.