def __init__(self,
input_size,
action_spec,
logits_init_output_factor=0.1,
name="CategoricalProjectionNetwork"):
"""Creates a categorical projection network that outputs a discrete
distribution over a number of classes.
Currently there seems no need for this class to handle nested inputs;
If necessary, extend the argument list to support it in the future.
Args:
input_size (int): the input vector size
action_spec (BounedTensorSpec): a tensor spec containing the information
of the output distribution.
name (str):
"""
super(CategoricalProjectionNetwork,
self).__init__(input_tensor_spec=TensorSpec((input_size, )),
name=name)
unique_num_actions = np.unique(action_spec.maximum -
action_spec.minimum + 1)
if len(unique_num_actions) > 1 or np.any(unique_num_actions <= 0):
raise ValueError(
'Bounds on discrete actions must be the same for all '
'dimensions and have at least 1 action. Projection '
'Network requires num_actions to be equal across '
'action dimensions. Implement a more general '
'categorical projection if you need more flexibility.')
output_shape = action_spec.shape + (int(unique_num_actions), )
self._output_shape = output_shape
self._projection_layer = layers.FC(
input_size,
np.prod(output_shape),
kernel_init_gain=logits_init_output_factor)
def __init__(self,
input_size,
action_spec,
activation=math_ops.identity,
projection_output_init_gain=0.3,
std_bias_initializer_value=0.0,
squash_mean=True,
state_dependent_std=False,
std_transform=nn.functional.softplus,
scale_distribution=False,
dist_squashing_transform=dist_utils.StableTanh(),
name="NormalProjectionNetwork"):
"""Creates an instance of NormalProjectionNetwork.
Currently there seems no need for this class to handle nested inputs;
If necessary, extend the argument list to support it in the future.
Args:
input_size (int): input vector dimension
action_spec (TensorSpec): a tensor spec containing the information
of the output distribution.
activation (Callable): activation function to use in
dense layers.
projection_output_init_gain (float): Output gain for initializing
action means and std weights.
std_bias_initializer_value (float): Initial value for the bias of the
``std_projection_layer``.
squash_mean (bool): If True, squash the output mean to fit the
action spec. If ``scale_distribution`` is also True, this value
will be ignored.
state_dependent_std (bool): If True, std will be generated depending
on the current state; otherwise a global std will be generated
regardless of the current state.
std_transform (Callable): Transform to apply to the std, on top of
`activation`.
scale_distribution (bool): Whether or not to scale the output
distribution to ensure that the output aciton fits within the
`action_spec`. Note that this is different from `mean_transform`
which merely squashes the mean to fit within the spec.
dist_squashing_transform (td.Transform): A distribution Transform
which transforms values into :math:`(-1, 1)`. Default to ``dist_utils.StableTanh()``
name (str): name of this network.
"""
super(NormalProjectionNetwork,
self).__init__(input_tensor_spec=TensorSpec((input_size, )),
name=name)
assert isinstance(action_spec, TensorSpec)
assert len(action_spec.shape) == 1, "Only support 1D action spec!"
self._action_spec = action_spec
self._mean_transform = math_ops.identity
self._scale_distribution = scale_distribution
if squash_mean or scale_distribution:
assert isinstance(action_spec, BoundedTensorSpec), \
("When squashing the mean or scaling the distribution, bounds "
+ "are required for the action spec!")
action_high = torch.as_tensor(action_spec.maximum)
action_low = torch.as_tensor(action_spec.minimum)
self._action_means = (action_high + action_low) / 2
self._action_magnitudes = (action_high - action_low) / 2
# Do not transform mean if scaling distribution
if not scale_distribution:
self._mean_transform = (
lambda inputs: self._action_means + self._action_magnitudes
* inputs.tanh())
else:
self._transforms = [
dist_squashing_transform,
dist_utils.AffineTransform(loc=self._action_means,
scale=self._action_magnitudes)
]
self._std_transform = math_ops.identity
if std_transform is not None:
self._std_transform = std_transform
self._means_projection_layer = layers.FC(
input_size,
action_spec.shape[0],
activation=activation,
kernel_init_gain=projection_output_init_gain)
if state_dependent_std:
self._std_projection_layer = layers.FC(
input_size,
action_spec.shape[0],
activation=activation,
kernel_init_gain=projection_output_init_gain,
bias_init_value=std_bias_initializer_value)
else:
self._std = nn.Parameter(
action_spec.constant(std_bias_initializer_value),
requires_grad=True)
self._std_projection_layer = lambda _: self._std
def __init__(self,
input_tensor_spec: TensorSpec,
action_spec: BoundedTensorSpec,
input_preprocessors=None,
preprocessing_combiner=None,
conv_layer_params=None,
fc_layer_params=None,
activation=torch.relu_,
kernel_initializer=None,
name="QNetwork"):
"""Creates an instance of ``QNetwork`` for estimating action-value of
discrete actions. The action-value is defined as the expected return
starting from the given input observation and taking the given action.
It takes observation as input and outputs an action-value tensor with
the shape of ``[batch_size, num_of_actions]``.
Args:
input_tensor_spec (TensorSpec): the tensor spec of the input
action_spec (TensorSpec): the tensor spec of the action
input_preprocessors (nested InputPreprocessor): a nest of
``InputPreprocessor``, each of which will be applied to the
corresponding input. If not None, then it must
have the same structure with ``input_tensor_spec`` (after reshaping).
If any element is None, then it will be treated as ``math_ops.identity``.
This arg is helpful if you want to have separate preprocessings
for different inputs by configuring a gin file without changing
the code. For example, embedding a discrete input before concatenating
it to another continuous vector.
preprocessing_combiner (NestCombiner): preprocessing called on
complex inputs. Note that this combiner must also accept
``input_tensor_spec`` as the input to compute the processed
tensor spec. For example, see ``alf.nest.utils.NestConcat``. This
arg is helpful if you want to combine inputs by configuring a
gin file without changing the code.
conv_layer_params (tuple[tuple]): a tuple of tuples where each
tuple takes a format ``(filters, kernel_size, strides, padding)``,
where ``padding`` is optional.
fc_layer_params (tuple[int]): a tuple of integers representing hidden
FC layer sizes.
activation (nn.functional): activation used for hidden layers. The
last layer will not be activated.
kernel_initializer (Callable): initializer for all the layers but
the last layer. If none is provided a default ``variance_scaling_initializer``
will be used.
"""
super(QNetwork, self).__init__(input_tensor_spec, name=name)
assert len(nest.flatten(action_spec)) == 1, (
"Currently only support a single discrete action! Use "
"CriticNetwork instead for multiple actions.")
num_actions = action_spec.maximum - action_spec.minimum + 1
self._output_spec = TensorSpec((num_actions, ))
self._encoding_net = EncodingNetwork(
input_tensor_spec=input_tensor_spec,
input_preprocessors=input_preprocessors,
preprocessing_combiner=preprocessing_combiner,
conv_layer_params=conv_layer_params,
fc_layer_params=fc_layer_params,
activation=activation,
kernel_initializer=kernel_initializer)
last_kernel_initializer = functools.partial(torch.nn.init.uniform_, \
a=-0.003, b=0.003)
self._final_layer = layers.FC(
self._encoding_net.output_spec.shape[0],
num_actions,
activation=math_ops.identity,
kernel_initializer=last_kernel_initializer,
bias_init_value=-0.2)
def __init__(self,
input_tensor_spec,
output_tensor_spec=None,
input_preprocessors=None,
preprocessing_combiner=None,
conv_layer_params=None,
fc_layer_params=None,
activation=torch.relu_,
kernel_initializer=None,
use_fc_bn=False,
last_layer_size=None,
last_activation=None,
last_kernel_initializer=None,
last_use_fc_bn=False,
name="EncodingNetwork"):
"""
Args:
input_tensor_spec (nested TensorSpec): the (nested) tensor spec of
the input. If nested, then ``preprocessing_combiner`` must not be
None.
output_tensor_spec (None|TensorSpec): spec for the output. If None,
the output tensor spec will be assumed as
``TensorSpec((output_size, ))``, where ``output_size`` is
inferred from network output. Otherwise, the output tensor
spec will be ``output_tensor_spec`` and the network output
will be reshaped according to ``output_tensor_spec``.
Note that ``output_tensor_spec`` is only used for reshaping
the network outputs for interpretation purpose and is not used
for specifying any network layers.
input_preprocessors (nested InputPreprocessor): a nest of
``InputPreprocessor``, each of which will be applied to the
corresponding input. If not None, then it must have the same
structure with ``input_tensor_spec``. This arg is helpful if you
want to have separate preprocessings for different inputs by
configuring a gin file without changing the code. For example,
embedding a discrete input before concatenating it to another
continuous vector.
preprocessing_combiner (NestCombiner): preprocessing called on
complex inputs. Note that this combiner must also accept
``input_tensor_spec`` as the input to compute the processed
tensor spec. For example, see ``alf.nest.utils.NestConcat``. This
arg is helpful if you want to combine inputs by configuring a
gin file without changing the code.
conv_layer_params (tuple[tuple]): a tuple of tuples where each
tuple takes a format ``(filters, kernel_size, strides, padding)``,
where ``padding`` is optional.
fc_layer_params (tuple[int]): a tuple of integers
representing FC layer sizes.
activation (nn.functional): activation used for all the layers but
the last layer.
kernel_initializer (Callable): initializer for all the layers but
the last layer. If None, a variance_scaling_initializer will be
used.
use_fc_bn (bool): whether use Batch Normalization for fc layers.
last_layer_size (int): an optional size of an additional layer
appended at the very end. Note that if ``last_activation`` is
specified, ``last_layer_size`` has to be specified explicitly.
last_activation (nn.functional): activation function of the
additional layer specified by ``last_layer_size``. Note that if
``last_layer_size`` is not None, ``last_activation`` has to be
specified explicitly.
last_use_fc_bn (bool): whether use Batch Normalization for the last
fc layer.
last_kernel_initializer (Callable): initializer for the the
additional layer specified by ``last_layer_size``.
If None, it will be the same with ``kernel_initializer``. If
``last_layer_size`` is None, ``last_kernel_initializer`` will
not be used.
name (str):
"""
super().__init__(input_tensor_spec,
input_preprocessors,
preprocessing_combiner,
name=name)
if kernel_initializer is None:
kernel_initializer = functools.partial(
variance_scaling_init,
mode='fan_in',
distribution='truncated_normal',
nonlinearity=activation)
self._img_encoding_net = None
if conv_layer_params:
assert isinstance(conv_layer_params, tuple), \
"The input params {} should be tuple".format(conv_layer_params)
assert len(self._processed_input_tensor_spec.shape) == 3, \
"The input shape {} should be like (C,H,W)!".format(
self._processed_input_tensor_spec.shape)
input_channels, height, width = self._processed_input_tensor_spec.shape
self._img_encoding_net = ImageEncodingNetwork(
input_channels, (height, width),
conv_layer_params,
activation=activation,
kernel_initializer=kernel_initializer,
flatten_output=True)
input_size = self._img_encoding_net.output_spec.shape[0]
else:
assert self._processed_input_tensor_spec.ndim == 1, \
"The input shape {} should be like (N,)!".format(
self._processed_input_tensor_spec.shape)
input_size = self._processed_input_tensor_spec.shape[0]
self._fc_layers = nn.ModuleList()
if fc_layer_params is None:
fc_layer_params = []
else:
assert isinstance(fc_layer_params, tuple)
fc_layer_params = list(fc_layer_params)
for size in fc_layer_params:
self._fc_layers.append(
layers.FC(input_size,
size,
activation=activation,
use_bn=use_fc_bn,
kernel_initializer=kernel_initializer))
input_size = size
if last_layer_size is not None or last_activation is not None:
assert last_layer_size is not None and last_activation is not None, \
"Both last_layer_size and last_activation need to be specified!"
if last_kernel_initializer is None:
common.warning_once(
"last_kernel_initializer is not specified "
"for the last layer of size {}.".format(last_layer_size))
last_kernel_initializer = kernel_initializer
self._fc_layers.append(
layers.FC(input_size,
last_layer_size,
activation=last_activation,
use_bn=last_use_fc_bn,
kernel_initializer=last_kernel_initializer))
input_size = last_layer_size
if output_tensor_spec is not None:
assert output_tensor_spec.numel == input_size, (
"network output "
"size {a} is inconsisent with specified out_tensor_spec "
"of size {b}".format(a=input_size, b=output_tensor_spec.numel))
self._output_spec = TensorSpec(
output_tensor_spec.shape,
dtype=self._processed_input_tensor_spec.dtype)
self._reshape_output = True
else:
self._output_spec = TensorSpec(
(input_size, ), dtype=self._processed_input_tensor_spec.dtype)
self._reshape_output = False
def __init__(self,
input_size,
transconv_layer_params,
start_decoding_size,
start_decoding_channels,
same_padding=False,
preprocess_fc_layer_params=None,
activation=torch.relu_,
kernel_initializer=None,
output_activation=torch.tanh,
name="ImageDecodingNetwork"):
"""
Initialize the layers for decoding a latent vector into an image.
Currently there seems no need for this class to handle nested inputs;
If necessary, extend the argument list to support it in the future.
How to calculate the output size:
`<https://pytorch.org/docs/stable/nn.html#torch.nn.ConvTranspose2d>`_::
H = (H1-1) * strides + HF - 2P + OP
where H = output size, H1 = input size, HF = size of kernel, P = padding,
OP = output_padding (currently hardcoded to be 0 for this class).
Regarding padding: in the previous TF version, we have two padding modes:
``valid`` and ``same``. For the former, we always have no padding (P=0); for
the latter, it's also called ``half padding`` (P=(HF-1)//2 when strides=1
and HF is an odd number the output has the same size with the input.
Currently, PyTorch doesn't support different left and right paddings and
P is always (HF-1)//2. So if HF is an even number, the output size will
increaseby 1 when strides=1).
Args:
input_size (int): the size of the input latent vector
transconv_layer_params (tuple[tuple]): a non-empty
tuple of tuple (num_filters, kernel_size, strides, padding),
where ``padding`` is optional.
start_decoding_size (int or tuple): the initial height and width
we'd like to have for the feature map
start_decoding_channels (int): the initial number of channels we'd
like to have for the feature map. Note that we always first
project an input latent vector into a vector of an appropriate
length so that it can be reshaped into (``start_decoding_channels``,
``start_decoding_height``, ``start_decoding_width``).
same_padding (bool): similar to TF's conv2d ``same`` padding mode. If
True, the user provided paddings in ``transconv_layer_params`` will
be replaced by automatically calculated ones; if False, it
corresponds to TF's ``valid`` padding mode (the user can still
provide custom paddings though).
preprocess_fc_layer_params (tuple[int]): a tuple of fc
layer units. These fc layers are used for preprocessing the
latent vector before transposed convolutions.
activation (nn.functional): activation for hidden layers
kernel_initializer (Callable): initializer for all the layers.
output_activation (nn.functional): activation for the output layer.
Usually our image inputs are normalized to [0, 1] or [-1, 1],
so this function should be ``torch.sigmoid`` or
``torch.tanh``.
name (str):
"""
super().__init__(input_tensor_spec=TensorSpec((input_size, )),
name=name)
assert isinstance(transconv_layer_params, tuple)
assert len(transconv_layer_params) > 0
self._preprocess_fc_layers = nn.ModuleList()
if preprocess_fc_layer_params is not None:
for size in preprocess_fc_layer_params:
self._preprocess_fc_layers.append(
layers.FC(input_size,
size,
activation=activation,
kernel_initializer=kernel_initializer))
input_size = size
start_decoding_size = common.tuplify2d(start_decoding_size)
# pytorch assumes "channels_first" !
self._start_decoding_shape = [
start_decoding_channels, start_decoding_size[0],
start_decoding_size[1]
]
self._preprocess_fc_layers.append(
layers.FC(input_size,
np.prod(self._start_decoding_shape),
activation=activation,
kernel_initializer=kernel_initializer))
self._transconv_layer_params = transconv_layer_params
self._transconv_layers = nn.ModuleList()
in_channels = start_decoding_channels
for i, paras in enumerate(transconv_layer_params):
filters, kernel_size, strides = paras[:3]
padding = paras[3] if len(paras) > 3 else 0
if same_padding: # overwrite paddings
kernel_size = common.tuplify2d(kernel_size)
padding = ((kernel_size[0] - 1) // 2,
(kernel_size[1] - 1) // 2)
act = activation
if i == len(transconv_layer_params) - 1:
act = output_activation
self._transconv_layers.append(
layers.ConvTranspose2D(in_channels,
filters,
kernel_size,
activation=act,
kernel_initializer=kernel_initializer,
strides=strides,
padding=padding))
in_channels = filters
def __init__(self,
input_tensor_spec: TensorSpec,
action_spec: BoundedTensorSpec,
input_preprocessors=None,
preprocessing_combiner=None,
conv_layer_params=None,
fc_layer_params=None,
activation=torch.relu_,
squashing_func=torch.tanh,
kernel_initializer=None,
name="ActorNetwork"):
"""Creates an instance of ``ActorNetwork``, which maps the inputs to
actions (single or nested) through a sequence of deterministic layers.
Args:
input_tensor_spec (TensorSpec): the tensor spec of the input.
action_spec (BoundedTensorSpec): the tensor spec of the action.
input_preprocessors (nested InputPreprocessor): a nest of
``InputPreprocessor``, each of which will be applied to the
corresponding input. If not None, then it must
have the same structure with ``input_tensor_spec`` (after reshaping).
If any element is None, then it will be treated as ``math_ops.identity``.
This arg is helpful if you want to have separate preprocessings
for different inputs by configuring a gin file without changing
the code. For example, embedding a discrete input before concatenating
it to another continuous vector.
preprocessing_combiner (NestCombiner): preprocessing called on
complex inputs. Note that this combiner must also accept
``input_tensor_spec`` as the input to compute the processed
tensor spec. For example, see ``alf.nest.utils.NestConcat``. This
arg is helpful if you want to combine inputs by configuring a
gin file without changing the code.
conv_layer_params (tuple[tuple]): a tuple of tuples where each
tuple takes a format ``(filters, kernel_size, strides, padding)``,
where ``padding`` is optional.
fc_layer_params (tuple[int]): a tuple of integers representing hidden
FC layer sizes.
activation (nn.functional): activation used for hidden layers. The
last layer will not be activated.
squashing_func (Callable): the activation function used to squashing
the output to the range :math:`(-1, 1)`. Default to ``tanh``.
kernel_initializer (Callable): initializer for all the layers but
the last layer. If none is provided a ``variance_scaling_initializer``
with uniform distribution will be used.
name (str): name of the network
"""
super(ActorNetwork, self).__init__(input_tensor_spec,
input_preprocessors,
preprocessing_combiner,
name=name)
if kernel_initializer is None:
kernel_initializer = functools.partial(variance_scaling_init,
gain=math.sqrt(1.0 / 3),
mode='fan_in',
distribution='uniform')
self._action_spec = action_spec
flat_action_spec = nest.flatten(action_spec)
self._flat_action_spec = flat_action_spec
is_continuous = [
single_action_spec.is_continuous
for single_action_spec in flat_action_spec
]
assert all(is_continuous), "only continuous action is supported"
self._encoding_net = EncodingNetwork(
input_tensor_spec=self._processed_input_tensor_spec,
conv_layer_params=conv_layer_params,
fc_layer_params=fc_layer_params,
activation=activation,
kernel_initializer=kernel_initializer,
name=self.name + ".encoding_net")
last_kernel_initializer = functools.partial(torch.nn.init.uniform_, \
a=-0.003, b=0.003)
self._action_layers = nn.ModuleList()
self._squashing_func = squashing_func
for single_action_spec in flat_action_spec:
self._action_layers.append(
layers.FC(self._encoding_net.output_spec.shape[0],
single_action_spec.shape[0],
kernel_initializer=last_kernel_initializer))