tower_2 = Conv2D(192, (1, 1), activation='relu')(model)
tower_2 = Conv2D(384, (3, 3), padding='same', activation='relu')(tower_2)
tower_3 = Conv2D(48, (1, 1), activation='relu')(model)
tower_3 = Conv2D(128, (5, 5), padding='same', activation='relu')(tower_3)
tower_4 = MaxPooling2D((3, 3), strides=1, padding='same')(model)
tower_4 = Conv2D(128, (1, 1), activation='relu')(tower_4)
model = Concatenate(axis=-1)([tower_1, tower_2, tower_3, tower_4])
model = AveragePooling2D((1, 1), strides=1, padding='valid')(model)
model = Flatten()(model)
model = Dropout(0.4)(model)
model = Dense(mnist_classes, activation='linear')(model)
output = Dense(mnist_classes, activation='softmax')(model)
model = Model(input, output)
model.summary()
"""# Implementation of Models on Datasets"""
import tensorflow as tf
from tensorflow.keras import datasets
import matplotlib.pyplot as plt
"""## MNIST"""
(train_images_mnist,
train_labels_mnist), (test_images_mnist,
test_labels_mnist) = datasets.mnist.load_data(
path='mnist.npz')
# Normalize pixel values to be between 0 and 1
train_images_mnist, test_images_mnist = train_images_mnist / 255.0, test_images_mnist / 255.0
def compileComponent(self,verbose=False,component_type=None):
if(component_type=='discriminator'):
stream = open(self.layout_dir+"/discriminator_layout.yaml","r+")
elif(component_type=='generator'):
stream = open(self.layout_dir+"/generator_layout.yaml","r+")
data = yaml.load(stream, Loader=yaml.FullLoader)
g1_instructions= data['join'][0]
g2_instructions= data['join'][1]
gc_instructions = data['layers']
if(component_type=='discriminator'):
g_in_1 = Input((1,))
elif(component_type=='generator'):
g_in_1 = Input((self.noise,))
g_in_2=[]
for i in range(self.dimensionality):
g_in_2.append(Input((1,)))
g1 = (g_in_1)
for layer_info in g1_instructions['layers']:
if(self.overrides.get('activation')):
activation = self.overrides.get('activation')
else:
activation = layer_info.get('activation')
if(component_type=='discriminator' and self.overrides.get('d_nodes')):
units = self.overrides.get('d_nodes')
elif(component_type=='generator' and self.overrides.get('g_nodes')):
units = self.overrides.get('g_nodes')
else:
units = layer_info.get('nodes')
if(self.overrides.get('dropout_amount')):
rate = self.overrides.get('dropout_amount')
else:
rate = layer_info.get('dropout_amount')
if(self.overrides.get('leaky_amount')):
alpha = self.overrides.get('leaky_amount')
else:
alpha = layer_info.get('leaky_amount')
if(layer_info['layer_type']=='dense'):
g1=Dense(units=units,activation=activation)(g1)
elif(layer_info['layer_type']=='dropout'):
g1=Dropout(rate=rate)(g1)
elif(layer_info['layer_type']=='selu'):
g1=LeakyReLU(alpha=alpha)(g1)
elif(layer_info['layer_type']=='batchnorm'):
g1=BatchNormalization()(g1)
g2=[]
for i in range(self.dimensionality):
g2_current = (g_in_2[i])
for layer_info in g2_instructions['layers']:
if(self.overrides.get('activation')):
activation = self.overrides.get('activation')
else:
activation = layer_info.get('activation')
if(component_type=='discriminator' and self.overrides.get('d_nodes')):
units = self.overrides.get('d_nodes')
elif(component_type=='generator' and self.overrides.get('g_nodes')):
units = self.overrides.get('g_nodes')
else:
units = layer_info.get('nodes')
if(self.overrides.get('dropout_amount')):
rate = self.overrides.get('dropout_amount')
else:
rate = layer_info.get('dropout_amount')
if(self.overrides.get('leaky_amount')):
alpha = self.overrides.get('leaky_amount')
else:
alpha = layer_info.get('leaky_amount')
if(layer_info['layer_type']=='dense'):
g2_current=Dense(units=units,activation=activation)(g2_current)
elif(layer_info['layer_type']=='dropout'):
g2_current=Dropout(rate=rate)(g2_current)
elif(layer_info['layer_type']=='selu'):
g2_current=LeakyReLU(alpha=alpha)(g2_current)
elif(layer_info['layer_type']=='batchnorm'):
g2_current=BatchNormalization()(g2_current)
g2.append(g2_current)
gc = Concatenate()([g1]+g2)
for layer_info in gc_instructions:
if(self.overrides.get('activation')):
activation = self.overrides.get('activation')
else:
activation = layer_info.get('activation')
if(component_type=='discriminator' and self.overrides.get('d_nodes')):
units = self.overrides.get('d_nodes')
elif(component_type=='generator' and self.overrides.get('g_nodes')):
units = self.overrides.get('g_nodes')
else:
units = layer_info.get('nodes')
if(self.overrides.get('dropout_amount')):
rate = self.overrides.get('dropout_amount')
else:
rate = layer_info.get('dropout_amount')
if(self.overrides.get('leaky_amount')):
alpha = self.overrides.get('leaky_amount')
else:
alpha = layer_info.get('leaky_amount')
if(layer_info['layer_type']=='dense'):
gc=Dense(units=units,activation=activation)(gc)
elif(layer_info['layer_type']=='dropout'):
gc=Dropout(rate=rate)(gc)
elif(layer_info['layer_type']=='selu'):
gc=LeakyReLU(alpha=alpha)(gc)
elif(layer_info['layer_type']=='batchnorm'):
gc=BatchNormalization()(gc)
if(component_type=='generator'):
gc = Dense(1, activation="sigmoid")(gc)
gc = Model(name="Generator", inputs=[g_in_1]+g_in_2, outputs=[gc])
elif(component_type=='discriminator'):
gc = Dense(2, activation="softmax")(gc)
gc = Model(name="Discriminator", inputs=[g_in_1]+g_in_2, outputs=[gc])
gc.compile(loss="categorical_crossentropy", optimizer=Adam(self.d_training_rate, beta_1=self.d_beta1), metrics=["accuracy"])
if(verbose):
gc.summary()
return gc
