# Copyright 2018 Tensorforce Team. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
import tensorflow as tf
from tensorforce import TensorforceError, util
from tensorforce.core.layers import TransformationBase
[docs]class Rnn(TransformationBase):
"""
Recurrent neural network layer (specification key: `rnn`).
Args:
name (string): Layer name
(<span style="color:#00C000"><b>default</b></span>: internally chosen).
cell ('gru' | 'lstm'): The recurrent cell type
(<span style="color:#C00000"><b>required</b></span>).
size (int >= 0): Layer output size, 0 implies additionally removing the axis
(<span style="color:#C00000"><b>required</b></span>).
return_final_state (bool): Whether to return the final state instead of the per-step
outputs (<span style="color:#00C000"><b>default</b></span>: true).
bias (bool): Whether to add a trainable bias variable
(<span style="color:#00C000"><b>default</b></span>: false).
activation ('crelu' | 'elu' | 'leaky-relu' | 'none' | 'relu' | 'selu' | 'sigmoid' |
'softmax' | 'softplus' | 'softsign' | 'swish' | 'tanh'): Activation nonlinearity
(<span style="color:#00C000"><b>default</b></span>: none).
dropout (parameter, 0.0 <= float < 1.0): Dropout rate
(<span style="color:#00C000"><b>default</b></span>: 0.0).
is_trainable (bool): Whether layer variables are trainable
(<span style="color:#00C000"><b>default</b></span>: true).
input_spec (specification): Input tensor specification
(<span style="color:#00C000"><b>internal use</b></span>).
summary_labels ('all' | iter[string]): Labels of summaries to record
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
l2_regularization (float >= 0.0): Scalar controlling L2 regularization
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
kwargs: Additional arguments for Keras RNN layer, see
`TensorFlow docs <https://www.tensorflow.org/api_docs/python/tf/keras/layers>`__.
"""
def __init__(
self, name, cell, size, return_final_state=True, bias=False, activation=None, dropout=0.0,
is_trainable=True, input_spec=None, summary_labels=None, l2_regularization=None, **kwargs
):
self.cell = cell
self.return_final_state = return_final_state
if self.return_final_state and self.cell == 'lstm':
assert size % 2 == 0
self.size = size // 2
else:
self.size = size
if self.cell == 'gru':
self.rnn = tf.keras.layers.GRU(
units=self.size, return_sequences=True, return_state=True, name='rnn',
input_shape=input_spec['shape'], **kwargs # , dtype=util.tf_dtype(dtype='float')
)
elif self.cell == 'lstm':
self.rnn = tf.keras.layers.LSTM(
units=self.size, return_sequences=True, return_state=True, name='rnn',
input_shape=input_spec['shape'], **kwargs # , dtype=util.tf_dtype(dtype='float')
)
else:
raise TensorforceError.unexpected()
super().__init__(
name=name, size=size, bias=bias, activation=activation, dropout=dropout,
is_trainable=is_trainable, input_spec=input_spec, summary_labels=summary_labels,
l2_regularization=l2_regularization
)
if self.squeeze and self.return_final_state:
raise TensorforceError(
"Invalid combination for Lstm layer: size=0 and return_final_state=True."
)
def default_input_spec(self):
return dict(type='float', shape=(-1, 0))
def get_output_spec(self, input_spec):
if self.return_final_state:
input_spec['shape'] = input_spec['shape'][:-2] + (self.size,)
elif self.squeeze:
input_spec['shape'] = input_spec['shape'][:-1]
else:
input_spec['shape'] = input_spec['shape'][:-1] + (self.size,)
input_spec.pop('min_value', None)
input_spec.pop('max_value', None)
return input_spec
def tf_initialize(self):
super().tf_initialize()
self.rnn.build(input_shape=((None,) + self.input_spec['shape']))
for variable in self.rnn.trainable_weights:
name = variable.name[variable.name.rindex(self.name + '/') + len(self.name) + 1: -2]
self.variables[name] = variable
if self.is_trainable:
self.trainable_variables[name] = variable
for variable in self.rnn.non_trainable_weights:
name = variable.name[variable.name.rindex(self.name + '/') + len(self.name) + 1: -2]
self.variables[name] = variable
def tf_regularize(self):
regularization_loss = super().tf_regularize()
if len(self.rnn.losses) > 0:
regularization_loss += tf.math.add_n(inputs=self.rnn.losses)
return regularization_loss
def tf_apply(self, x, sequence_length=None):
x = self.rnn(inputs=x, initial_state=None)
if self.return_final_state:
if self.cell == 'gru':
x = x[1]
elif self.cell == 'lstm':
x = tf.concat(values=(x[1], x[2]), axis=1)
else:
x = x[0]
return super().tf_apply(x=x)
[docs]class Gru(Rnn):
"""
Gated recurrent unit layer (specification key: `gru`).
Args:
name (string): Layer name
(<span style="color:#00C000"><b>default</b></span>: internally chosen).
cell ('gru' | 'lstm'): The recurrent cell type
(<span style="color:#C00000"><b>required</b></span>).
size (int >= 0): Layer output size, 0 implies additionally removing the axis
(<span style="color:#C00000"><b>required</b></span>).
return_final_state (bool): Whether to return the final state instead of the per-step
outputs (<span style="color:#00C000"><b>default</b></span>: true).
bias (bool): Whether to add a trainable bias variable
(<span style="color:#00C000"><b>default</b></span>: false).
activation ('crelu' | 'elu' | 'leaky-relu' | 'none' | 'relu' | 'selu' | 'sigmoid' |
'softmax' | 'softplus' | 'softsign' | 'swish' | 'tanh'): Activation nonlinearity
(<span style="color:#00C000"><b>default</b></span>: none).
dropout (parameter, 0.0 <= float < 1.0): Dropout rate
(<span style="color:#00C000"><b>default</b></span>: 0.0).
is_trainable (bool): Whether layer variables are trainable
(<span style="color:#00C000"><b>default</b></span>: true).
input_spec (specification): Input tensor specification
(<span style="color:#00C000"><b>internal use</b></span>).
summary_labels ('all' | iter[string]): Labels of summaries to record
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
l2_regularization (float >= 0.0): Scalar controlling L2 regularization
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
kwargs: Additional arguments for Keras GRU layer, see
`TensorFlow docs <https://www.tensorflow.org/api_docs/python/tf/keras/layers/GRU>`__.
"""
def __init__(
self, name, size, return_final_state=True, bias=False, activation=None, dropout=0.0,
is_trainable=True, input_spec=None, summary_labels=None, l2_regularization=None, **kwargs
):
super().__init__(
name=name, cell='gru', size=size, return_final_state=return_final_state, bias=bias,
activation=activation, dropout=dropout, input_spec=input_spec,
summary_labels=summary_labels, l2_regularization=l2_regularization, **kwargs
)
[docs]class Lstm(Rnn):
"""
Long short-term memory layer (specification key: `lstm`).
Args:
name (string): Layer name
(<span style="color:#00C000"><b>default</b></span>: internally chosen).
cell ('gru' | 'lstm'): The recurrent cell type
(<span style="color:#C00000"><b>required</b></span>).
size (int >= 0): Layer output size, 0 implies additionally removing the axis
(<span style="color:#C00000"><b>required</b></span>).
return_final_state (bool): Whether to return the final state instead of the per-step
outputs (<span style="color:#00C000"><b>default</b></span>: true).
bias (bool): Whether to add a trainable bias variable
(<span style="color:#00C000"><b>default</b></span>: false).
activation ('crelu' | 'elu' | 'leaky-relu' | 'none' | 'relu' | 'selu' | 'sigmoid' |
'softmax' | 'softplus' | 'softsign' | 'swish' | 'tanh'): Activation nonlinearity
(<span style="color:#00C000"><b>default</b></span>: none).
dropout (parameter, 0.0 <= float < 1.0): Dropout rate
(<span style="color:#00C000"><b>default</b></span>: 0.0).
is_trainable (bool): Whether layer variables are trainable
(<span style="color:#00C000"><b>default</b></span>: true).
input_spec (specification): Input tensor specification
(<span style="color:#00C000"><b>internal use</b></span>).
summary_labels ('all' | iter[string]): Labels of summaries to record
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
l2_regularization (float >= 0.0): Scalar controlling L2 regularization
(<span style="color:#00C000"><b>default</b></span>: inherit value of parent module).
kwargs: Additional arguments for Keras LSTM layer, see
`TensorFlow docs <https://www.tensorflow.org/api_docs/python/tf/keras/layers/LSTM>`__.
"""
def __init__(
self, name, size, return_final_state=True, bias=False, activation=None, dropout=0.0,
is_trainable=True, input_spec=None, summary_labels=None, l2_regularization=None, **kwargs
):
super().__init__(
name=name, cell='lstm', size=size, return_final_state=return_final_state, bias=bias,
activation=activation, dropout=dropout, input_spec=input_spec,
summary_labels=summary_labels, l2_regularization=l2_regularization, **kwargs
)