def EmbeddingNet_classification(n1_features,
n2_features,
n_users,
n_latent_factors_user=8,
n_latent_factors_item=8,
k=200,
alpha=0.15,
dropout=0.2,
lr=0.005):
# Embedding usera
model1_in = Input(shape=(n1_features, ), name='useraInput')
model1_out = Embedding(input_dim=n_users + 1,
output_dim=n_latent_factors_user)(model1_in)
model1_out = Flatten()(model1_out)
model1_out = Dropout(dropout)(model1_out)
# Embedding userb
model2_in = Input(shape=(n2_features, ), name='userbInput')
model2_out = Embedding(input_dim=n_users + 1,
output_dim=n_latent_factors_item)(model2_in)
model2_out = Flatten()(model2_out)
model2_out = Dropout(dropout)(model2_out)
# Merge embedding of usera and embeddingof userb
model = concatenate([model1_out, model2_out], axis=-1)
model = LeakyReLU(alpha=alpha)(model)
model = Dropout(dropout)(model)
# Deep Learning
model = Dense(k)(model)
model = LeakyReLU(alpha=alpha)(model)
model = Dropout(dropout)(model)
model = Dense(int(k / 2))(model)
model = LeakyReLU(alpha=alpha)(model)
model = Dropout(dropout)(model)
model = Dense(int(k / 4))(model)
model = LeakyReLU(alpha=alpha)(model)
model = Dropout(dropout)(model)
model = Dense(int(k / 10))(model)
model = LeakyReLU(alpha=alpha)(model)
model = Dense(2, activation='softmax')(model)
model = Model([model1_in, model2_in], model)
adam = Adam(lr=lr)
model.compile(optimizer=adam,
loss='categorical_crossentropy',
metrics=['accuracy'])
return model
def create_discriminator():
discriminator_input = Input(shape=(28, 28, 1))
discriminator = Conv2D(65, 5, strides=(2, 2), padding='same')(discriminator_input)
discriminator = LeakyReLU()(discriminator)
discriminator = Dropout(0.4)(discriminator)
discriminator = Conv2D(256, 5, strides=(2, 2), padding='same')(discriminator)
discriminator = LeakyReLU()(discriminator)
discriminator = Dropout(0.4)(discriminator)
discriminator = Conv2D(512, 5, strides=(2, 2), padding='same')(discriminator)
discriminator = LeakyReLU()(discriminator)
discriminator = Dropout(0.4)(discriminator)
discriminator = Flatten()(discriminator)
discriminator_output = Dense(1, activation='sigmoid')(discriminator)
discriminator = Model(discriminator_input, discriminator_output)
discriminator.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
discriminator.summary()
return discriminator
def EmbeddingNetV3(n1_features,
n2_features,
n_latent_factors_user=5,
n_latent_factors_item=8,
n_users=743):
model1_in = Input(shape=(n1_features, ), name='userInput')
model1_out = Embedding(input_dim=n_users + 1,
output_dim=n_latent_factors_user)(model1_in)
model1_out = Flatten()(model1_out)
model1_out = Dropout(0.2)(model1_out)
model2_in = Input(shape=(n2_features, ), name='itemInput')
model2_out = Embedding(input_dim=n_users + 1,
output_dim=n_latent_factors_item)(model2_in)
model2_out = Flatten()(model2_out)
model2_out = Dropout(0.2)(model2_out)
model = concatenate([model1_out, model2_out], axis=-1)
model = LeakyReLU(alpha=0.15)(model)
model = Dropout(0.2)(model)
model = Dense(200)(model)
model = LeakyReLU(alpha=0.15)(model)
model = Dropout(0.2)(model)
model = Dense(100)(model)
model = LeakyReLU(alpha=0.15)(model)
model = Dropout(0.2)(model)
model = Dense(50)(model)
model = LeakyReLU(alpha=0.15)(model)
model = Dropout(0.2)(model)
model = Dense(20)(model)
model = LeakyReLU(alpha=0.15)(model)
model = Dense(2, activation='softmax')(model)
adam = Adam(lr=0.005)
model = Model([model1_in, model2_in], model)
model.compile(optimizer=adam,
loss='categorical_crossentropy',
metrics=['accuracy'])
return model
def build_discriminator(self):
# Source image input
source_img_in = Input(shape = self.image_shape)
# Target image input
target_img_in = Input(shape = self.image_shape)
# concatenate images
merged = Concatenate()([source_img_in, target_img_in])
# C64
model = Conv2D(64, (4,4), strides=(2,2), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(merged)
model = LeakyReLU(alpha=0.2)(model)
# C128
model = Conv2D(128, (4,4), strides=(2,2), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(model)
model = BatchNormalization()(model)
model = LeakyReLU(alpha=0.2)(model)
# C256
model = Conv2D(256, (4,4), strides=(2,2), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(model)
model = BatchNormalization()(model)
model = LeakyReLU(alpha=0.2)(model)
# C512
model = Conv2D(512, (4,4), strides=(2,2), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(model)
model = BatchNormalization()(model)
model = LeakyReLU(alpha=0.2)(model)
# Last layer
model = Conv2D(256, (4,4), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(model)
model = BatchNormalization()(model)
model = LeakyReLU(alpha=0.2)(model)
# Patch output
model = Conv2D(1, (4,4), padding = "same", kernel_initializer = RandomNormal(stddev=0.02))(model)
patch_out = Activation("sigmoid")(model)
# Define model
model = Model([source_img_in, target_img_in], patch_out)
# Compile model
opt = Adam(lr=0.0002, beta_2=0.5)
model.compile(loss = "binary_crossentropy", optimizer=opt, loss_weights=[0.5])
return model
