def summary(self, line_length=None, positions=None, print_fn=None):
"""Prints a string summary of the network.
Args:
line_length: Total length of printed lines
(e.g. set this to adapt the display to different
terminal window sizes).
positions: Relative or absolute positions of log elements
in each line. If not provided,
defaults to `[.33, .55, .67, 1.]`.
print_fn: Print function to use. Defaults to `print`.
It will be called on each line of the summary.
You can set it to a custom function
in order to capture the string summary.
Raises:
ValueError: if `summary()` is called before the model is built.
"""
if not self.built:
raise ValueError(
"This model has not yet been built. "
"Build the model first by calling `build()` or "
"`__call__()` with some data, or `create_variables()`.")
layer_utils.print_summary(self,
line_length=line_length,
positions=positions,
print_fn=print_fn)
def build(self, inputs_shape):
""" Build TCN architecture
The `inputs_shape` argument is a `(N, T, D)` tuple where `N` denotes the number of samples, `T` the number of
time-frames, and `D` the number of channels
"""
x = tf.keras.layers.Input(inputs_shape[1:])
dropout_rate = 1 - self.config.keep_prob
net = x
for _ in range(self.config.nb_conv_stacks):
net = tf.keras.layers.Conv1D(
filters=self.config.nb_conv_filters,
kernel_size=self.config.kernel_size,
strides=self.config.nb_conv_strides,
padding=self.config.padding,
kernel_initializer=self.config.kernel_initializer,
kernel_regularizer=tf.keras.regularizers.l2(
self.config.kernel_regularizer))(net)
if self.config.use_batch_norm:
net = tf.keras.layers.BatchNormalization(axis=-1)(net)
net = tf.keras.layers.Activation(self.config.activation)(net)
net = tf.keras.layers.Dropout(dropout_rate)(net)
net = tf.keras.layers.Flatten()(net)
for _ in range(self.config.nb_fc_stacks):
net = tf.keras.layers.Dense(
units=self.config.nb_fc_neurons,
kernel_initializer=self.config.kernel_initializer,
kernel_regularizer=tf.keras.regularizers.l2(
self.config.kernel_regularizer))(net)
if self.config.use_batch_norm:
net = tf.keras.layers.BatchNormalization(axis=-1)(net)
net = tf.keras.layers.Activation(self.config.activation)(net)
net = tf.keras.layers.Dropout(dropout_rate)(net)
net = tf.keras.layers.Dense(
units=self.config.n_classes,
kernel_initializer=self.config.kernel_initializer,
kernel_regularizer=tf.keras.regularizers.l2(
self.config.kernel_regularizer))(net)
net = tf.keras.layers.Softmax()(net)
self.net = tf.keras.Model(inputs=x, outputs=net)
print_summary(self.net)
def build(self, inputs_shape):
""" Build Transformer encoder architecture
The `inputs_shape` argument is a `(N, T, D)` tuple where `N` denotes the number of samples, `T` the number of
time-frames, and `D` the number of channels
"""
seq_len = inputs_shape[1]
self.net = tf.keras.Sequential([
self.encoder, self.dense,
tf.keras.layers.MaxPool1D(pool_size=seq_len),
tf.keras.layers.Lambda(
lambda x: tf.keras.backend.squeeze(x, axis=-2),
name='squeeze'),
tf.keras.layers.Softmax()
])
# Build the model, so we can print the summary
self.net.build(inputs_shape)
print_summary(self.net)
def main():
features, labels = loadFromPickle()
# features, labels = augmentData(features, labels)
features, labels = shuffle(features, labels)
labels = prepress_labels(labels)
train_x, test_x, train_y, test_y = train_test_split(features,
labels,
random_state=0,
test_size=0.1)
train_x = train_x.reshape(train_x.shape[0], 28, 28, 1)
test_x = test_x.reshape(test_x.shape[0], 28, 28, 1)
model, callbacks_list = keras_model(28, 28)
print_summary(model)
model.fit(train_x,
train_y,
validation_data=(test_x, test_y),
epochs=50,
batch_size=64,
callbacks=[TensorBoard(log_dir="QuickDraw")])
model.save('QuickDraw.h5')
def build(self, inputs_shape):
""" Build TCN architecture
The `inputs_shape` argument is a `(N, T, D)` tuple where `N` denotes the number of samples, `T` the number of
time-frames, and `D` the number of channels
"""
x = tf.keras.layers.Input(inputs_shape[1:])
dropout_rate = 1 - self.config.keep_prob
net = x
net = tf.keras.layers.Conv1D(
filters=self.config.nb_filters,
kernel_size=1,
padding=self.config.padding,
kernel_initializer=self.config.kernel_initializer)(net)
# list to hold all the member ResidualBlocks
residual_blocks = list()
skip_connections = list()
total_num_blocks = self.config.nb_stacks * len(self.config.dilations)
if not self.config.use_skip_connections:
total_num_blocks += 1 # cheap way to do a false case for below
for s in range(self.config.nb_stacks):
for d in self.config.dilations:
net, skip_out = ResidualBlock(
dilation_rate=d,
nb_filters=self.config.nb_filters,
kernel_size=self.config.kernel_size,
padding=self.config.padding,
activation=self.config.activation,
dropout_rate=dropout_rate,
use_batch_norm=self.config.use_batch_norm,
use_layer_norm=self.config.use_layer_norm,
kernel_initializer=self.config.kernel_initializer,
last_block=len(residual_blocks) + 1 == total_num_blocks,
name=f'residual_block_{len(residual_blocks)}')(net)
residual_blocks.append(net)
skip_connections.append(skip_out)
# Author: @karolbadowski.
output_slice_index = int(net.shape.as_list()[1] / 2) \
if self.config.padding.lower() == 'same' else -1
lambda_layer = tf.keras.layers.Lambda(
lambda tt: tt[:, output_slice_index, :])
if self.config.use_skip_connections:
net = tf.keras.layers.add(skip_connections)
if not self.config.return_sequences:
net = lambda_layer(net)
net = tf.keras.layers.Dense(self.config.n_classes)(net)
net = tf.keras.layers.Softmax()(net)
self.net = tf.keras.Model(inputs=x, outputs=net)
print_summary(self.net)
def build(self, inputs_shape):
self.net.build(inputs_shape)
print_summary(self.net)