def model(th):
assert isinstance(th, m.Config)
model = m.get_container(th, flatten=True)
model.add(Dense(th.layer_width, th.spatial_activation))
model.add(Highway(
th.layer_width, th.num_layers, th.spatial_activation,
t_bias_initializer=th.bias_initializer))
# model.register_extractor(m.LinearHighway.extractor)
return m.finalize(th, model)
def output_and_build(model, th):
assert isinstance(model, Classifier)
assert isinstance(th, Config)
# Add dropout if necessary
if th.output_dropout > 0: model.add(Dropout(1 - th.output_dropout))
# Add output layer
model.add(Dense(num_neurons=th.output_dim))
model.add(Activation('softmax'))
model.build(loss='cross_entropy', metric='bpc', batch_metric='bpc')
def output_and_build(model, th):
assert isinstance(model, Classifier)
assert isinstance(th, Config)
# Add output layer
model.add(Dense(num_neurons=th.output_dim))
model.add(Activation('softmax'))
model.build(metric='gen_acc',
batch_metric='gen_acc',
val_targets='val_targets')
def output_and_build(model, th):
assert isinstance(model, Classifier) and isinstance(th, Config)
# Add output dropout if necessary
if th.output_dropout > 0: model.add(Dropout(1 - th.output_dropout))
# Add output layer
model.add(Dense(num_neurons=th.output_dim))
model.add(Activation('softmax'))
# Build model
model.build(loss=th.loss_string, metric=['loss', 'f1'], batch_metric='f1')
def _get_layers(self):
layers = []
# Add conv layers
for n in self.conv_list: layers.append(Conv2D(
n, self.kernel_size, strides=self.strides,
padding=self.padding, activation=self.activation))
# Add flatten layer
layers.append(Flatten())
# Add fully-connected layers
for n in self.fc_list: layers.append(Dense(n ,activation=self.activation))
return layers
def output_and_build(model, th):
assert isinstance(model, Classifier)
assert isinstance(th, Config)
# Add dropout if necessary
if th.output_dropout > 0: model.add(Dropout(1 - th.output_dropout))
# Add output layer
model.add(Dense(num_neurons=th.output_dim))
model.add(Activation('softmax'))
model.build(last_only=True,
metric=['accuracy', 'loss'],
batch_metric='accuracy',
eval_metric='accuracy')
def model(th):
assert isinstance(th, m.Config)
model = m.get_container(th, flatten=True)
model.add(Dense(th.layer_width, th.spatial_activation))
model.add(SLHighway(
config_string=th.group_string,
num_layers=th.num_layers,
head_size=th.head_size,
activation=th.spatial_activation,
gutter=th.gutter,
gutter_bias=th.gutter_bias,
))
model.register_extractor(SLHighway.extractor)
return m.finalize(th, model)
def typical(th, cells):
assert isinstance(th, Config)
# Initiate a model
model = Classifier(mark=th.mark, net_type=Recurrent)
# Add layers
model.add(Input(sample_shape=th.input_shape, dtype=tf.int32))
model.add(Onehot(depth=th.num_classes))
emb_init = tf.initializers.random_uniform(-1, 1)
model.add(Dense(th.hidden_dim, use_bias=False,
weight_initializer=emb_init))
if th.input_dropout > 0: model.add(Dropout(1 - th.input_dropout))
# Add hidden layers
if not isinstance(cells, (list, tuple)): cells = [cells]
for cell in cells:
model.add(cell)
# Build model and return
output_and_build(model, th)
return model
