def __init__(self,
size,
window=3,
stride=1,
padding='SAME',
bias=True,
activation='relu',
l2_regularization=0.0,
l1_regularization=0.0,
scope='conv1d',
summary_labels=()):
"""
1D convolutional layer.
Args:
size: Number of filters
window: Convolution window size
stride: Convolution stride
padding: Convolution padding, one of 'VALID' or 'SAME'
bias: If true, a bias is added
activation: Type of nonlinearity, or dict with name & arguments
l2_regularization: L2 regularization weight
l1_regularization: L1 regularization weight
"""
self.size = size
self.window = window
self.stride = stride
self.padding = padding
self.bias = bias
self.l2_regularization = l2_regularization
self.l1_regularization = l1_regularization
self.nonlinearity = Nonlinearity(summary_labels=summary_labels,
**util.prepare_kwargs(activation))
super(Conv1d, self).__init__(scope=scope,
summary_labels=summary_labels)
def __init__(
self,
size=None,
weights=None,
bias=True,
activation='relu',
l2_regularization=0.0,
l1_regularization=0.0,
skip=False,
trainable=True,
named_tensors=None,
scope='dense',
summary_labels=(),
):
"""
Dense layer.
Args:
size: Layer size, if None than input size matches the output size of the layer
weights: Weight initialization, random if None.
bias: If true, bias is added.
activation: Type of nonlinearity, or dict with name & arguments
l2_regularization: L2 regularization weight.
l1_regularization: L1 regularization weight.
skip: Add skip connection like ResNet (https://arxiv.org/pdf/1512.03385.pdf),
doubles layers and ShortCut from Input to output
"""
self.skip = skip
if self.skip and size is not None:
raise TensorForceError(
'Dense Layer SKIP connection needs Size=None, uses input shape '
'sizes to create skip connection network, please delete "size" parameter'
)
self.linear = Linear(size=size,
weights=weights,
bias=bias,
l2_regularization=l2_regularization,
l1_regularization=l1_regularization,
summary_labels=summary_labels,
trainable=trainable)
if self.skip:
self.linear_skip = Linear(size=size,
bias=bias,
l2_regularization=l2_regularization,
l1_regularization=l1_regularization,
summary_labels=summary_labels,
trainable=trainable)
# TODO: Consider creating two nonlinearity variables when skip is used and learning beta
# Right now, only a single beta can be learned
self.nonlinearity = Nonlinearity(summary_labels=summary_labels,
**util.prepare_kwargs(activation))
super(Dense, self).__init__(named_tensors=named_tensors,
scope=scope,
summary_labels=summary_labels)
def __init__(self,
size,
bias=False,
activation='none',
l2_regularization=0.0,
l1_regularization=0.0,
output=None,
named_tensors=None,
scope='dueling',
summary_labels=()):
"""
Dueling layer.
[Dueling Networks] (https://arxiv.org/pdf/1511.06581.pdf)
Implement Y = Expectation[x] + (Advantage[x] - Mean(Advantage[x]))
Args:
size: Layer size.
bias: If true, bias is added.
activation: Type of nonlinearity, or dict with name & arguments
l2_regularization: L2 regularization weight.
l1_regularization: L1 regularization weight.
output: None or tuple of output names for ('expectation','advantage','mean_advantage')
"""
# Expectation is broadcast back over advantage values so output is of size 1
self.expectation_layer = Linear(
size=1,
bias=bias,
l2_regularization=l2_regularization,
l1_regularization=l1_regularization,
summary_labels=summary_labels,
)
self.advantage_layer = Linear(
size=size,
bias=bias,
l2_regularization=l2_regularization,
l1_regularization=l1_regularization,
summary_labels=summary_labels,
)
self.output = output
self.nonlinearity = Nonlinearity(summary_labels=summary_labels,
**util.prepare_kwargs(activation))
super(Dueling, self).__init__(named_tensors=named_tensors,
scope=scope,
summary_labels=summary_labels)
def __init__(self,
size,
window=3,
stride=1,
padding='SAME',
bias=True,
activation='relu',
l2_regularization=0.0,
l1_regularization=0.0,
named_tensors=None,
scope='conv2d',
summary_labels=()):
"""
2D convolutional layer.
Args:
size: Number of filters
window: Convolution window size, either an integer or pair of integers.
stride: Convolution stride, either an integer or pair of integers.
padding: Convolution padding, one of 'VALID' or 'SAME'
bias: If true, a bias is added
activation: Type of nonlinearity, or dict with name & arguments
l2_regularization: L2 regularization weight
l1_regularization: L1 regularization weight
"""
self.size = size
if isinstance(window, int):
self.window = (window, window)
elif len(window) == 2:
self.window = tuple(window)
else:
raise TensorForceError(
'Invalid window {} for conv2d layer, must be of size 2'.format(
window))
self.stride = stride
self.padding = padding
self.bias = bias
self.l2_regularization = l2_regularization
self.l1_regularization = l1_regularization
self.nonlinearity = Nonlinearity(summary_labels=summary_labels,
**util.prepare_kwargs(activation))
super(Conv2d, self).__init__(named_tensors=named_tensors,
scope=scope,
summary_labels=summary_labels)
