def simple_model(features=16, layers=1, name=None):
"""Creates and returns a simple keras model
Fully connected model:
-Flatten board grid into vector
-`layers` # of Dense layers with `features` # of nodes
-Dense layer with 8 nodes
-Dense layer of size 4 using softmax for output
Args:
features: number of nodes for middle layers
Defaults to 16
layers: number of middle layers
Defaults to 1
name: optional name for model
"""
model = Sequential()
if name:
model.name = name
model.add(Flatten(input_shape=(SIZE, SIZE)))
# TODO: Test effect of batch norm
for _ in range(layers):
model.add(Dense(features, activation='relu'))
model.add(Dense(8, activation='relu'))
model.add(Dense(4, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam')
return model
def train_discriminator(y_pred_in, y_pred_out, validation_data,
featuremap_attacker):
if featuremap_attacker is None:
model = Sequential()
model.name = "featuremap_mia"
model.add(Dense(input_shape=(y_pred_in.shape[1:]), units=500))
model.add(Dropout(0.2))
model.add(Dense(units=250))
model.add(Dropout(0.2))
model.add(Dense(units=10))
model.add(Dense(units=1, activation="sigmoid"))
model.compile(optimizer="Adam",
metrics=["accuracy"],
loss="binary_crossentropy")
featuremap_attacker = model
featuremap_attacker.fit(np.concatenate((y_pred_in, y_pred_out),
axis=0),
np.concatenate((np.zeros(len(y_pred_in)),
np.ones(len(y_pred_out)))),
validation_data=validation_data,
epochs=epochs,
verbose=0)
return featuremap_attacker
def _create_one_layer_model(input_shape: Optional[tuple], actions_count: int,
units: int) -> Sequential:
model = Sequential()
model.name = "one-hidden-layer-" + str(units)
model.add(Dense(units=units, input_shape=input_shape, activation='relu'))
model.add(Dense(units=actions_count, activation='linear'))
model.compile(loss='mse', optimizer='adam')
return model
