def batchnorm_activation_dropout(input_layer, batchnorm, activation_function,
dropout_rate, is_training, name):
layers = [input_layer]
# batchnorm
if batchnorm:
layers.append(
tf.layers.batch_normalization(layers[-1],
name="{}_batchnorm".format(name),
training=is_training))
# activation
if activation_function:
if isinstance(activation_function, str):
activation_function = utils.get_activation_function(
activation_function)
layers.append(
activation_function(layers[-1], name="{}_activation".format(name)))
# dropout
if dropout_rate > 0:
layers.append(
tf.layers.dropout(layers[-1],
dropout_rate,
name="{}_dropout".format(name),
training=is_training))
# remove the input layer from the layers list
del layers[0]
return layers
def batchnorm_activation_dropout(input_layer, batchnorm, activation_function,
dropout_rate, is_training, name):
layers = [input_layer]
# Rationale: passing a bool here will mean that batchnorm and or activation will never activate
assert not isinstance(is_training, bool)
# batchnorm
if batchnorm:
layers.append(
tf.layers.batch_normalization(layers[-1],
name="{}_batchnorm".format(name),
training=is_training))
# activation
if activation_function:
if isinstance(activation_function, str):
activation_function = utils.get_activation_function(
activation_function)
layers.append(
activation_function(layers[-1], name="{}_activation".format(name)))
# dropout
if dropout_rate > 0:
layers.append(
tf.layers.dropout(layers[-1],
dropout_rate,
name="{}_dropout".format(name),
training=is_training))
# remove the input layer from the layers list
del layers[0]
return layers
