torchrecurrent.WMCLSTMCell

class torchrecurrent.WMCLSTMCell(input_size, hidden_size, bias=True, recurrent_bias=True, memory_bias=True, kernel_init=<function xavier_uniform_>, recurrent_kernel_init=<function xavier_uniform_>, memory_kernel_init=<function xavier_uniform_>, bias_init=<function zeros_>, recurrent_bias_init=<function zeros_>, memory_bias_init=<function zeros_>, device=None, dtype=None)

A long short-term memory cell with working memory connections (WMCLSTMCell).

Based on arXiv:2109.00020.

The cell update equations are:

\[\begin{split}\begin{aligned} \mathbf{i}(t) &= \sigma\bigl( \mathbf{W}_{ih}^{i}\,\mathbf{x}(t) + \mathbf{b}_{ih}^{i} + \mathbf{W}_{hh}^{i}\,\mathbf{h}(t-1) + \mathbf{b}_{hh}^{i} + \mathbf{W}_{mh}^{i}\,\mathbf{c}(t-1) + \mathbf{b}_{mh}^{i} \bigr), \\ \mathbf{f}(t) &= \sigma\bigl( \mathbf{W}_{ih}^{f}\,\mathbf{x}(t) + \mathbf{b}_{ih}^{f} + \mathbf{W}_{hh}^{f}\,\mathbf{h}(t-1) + \mathbf{b}_{hh}^{f} + \mathbf{W}_{mh}^{f}\,\mathbf{c}(t-1) + \mathbf{b}_{mh}^{f} \bigr), \\ \mathbf{c}(t) &= \mathbf{f}(t) \circ \mathbf{c}(t-1) \;+\;\mathbf{i}(t)\circ\sigma_c\bigl(\mathbf{W}_{ih}^{c}\,\mathbf{x}(t) + \mathbf{b}_{ih}^{c}\bigr), \\ \mathbf{o}(t) &= \sigma\bigl( \mathbf{W}_{ih}^{o}\,\mathbf{x}(t) + \mathbf{b}_{ih}^{o} + \mathbf{W}_{hh}^{o}\,\mathbf{h}(t-1) + \mathbf{b}_{hh}^{o} + \mathbf{W}_{mh}^{o}\,\mathbf{c}(t) + \mathbf{b}_{mh}^{o} \bigr), \\ \mathbf{h}(t) &= \mathbf{o}(t)\circ\sigma_h\bigl(\mathbf{c}(t)\bigr), \end{aligned}\end{split}\]

where \(\sigma\) is the sigmoid function, \(\sigma_c\) and \(\sigma_h\) are cell and output activations (here both are torch.tanh), and \(\circ\) denotes elementwise multiplication.

Parameters:

  • input_size (int) – Number of features in the input \(\mathbf{x}(t)\).

  • hidden_size (int) – Number of features in the hidden state \(\mathbf{h}(t)\).

  • bias (bool, optional) – If False, disables \(\mathbf{b}_{ih}\). Default: True.

  • recurrent_bias (bool, optional) – If False, disables \(\mathbf{b}_{hh}\). Default: True.

  • memory_bias (bool, optional) – If False, disables \(\mathbf{b}_{mh}\). Default: True.

  • kernel_init (Callable, optional) – Initializer for input-to-hidden weights \(\mathbf{W}_{ih}^{\{i,f,c,o\}}\). Default: nn.init.xavier_uniform_.

  • recurrent_kernel_init (Callable, optional) – Initializer for hidden-to-hidden weights \(\mathbf{W}_{hh}^{\{i,f,o\}}\). Default: nn.init.xavier_uniform_.

  • memory_kernel_init (Callable, optional) – Initializer for working-memory weights \(\mathbf{W}_{mh}^{\{i,f,o\}}\). Default: nn.init.xavier_uniform_.

  • bias_init (Callable, optional) – Initializer for input biases \(\mathbf{b}_{ih}^{\{i,f,c,o\}}\). Default: nn.init.zeros_.

  • recurrent_bias_init (Callable, optional) – Initializer for hidden biases \(\mathbf{b}_{hh}^{\{i,f,o\}}\). Default: nn.init.zeros_.

  • memory_bias_init (Callable, optional) – Initializer for memory biases \(\mathbf{b}_{mh}^{\{i,f,o\}}\). Default: nn.init.zeros_.

  • device (torch.device, optional) – Device for parameters.

  • dtype (torch.dtype, optional) – Data type for parameters.

Inputs:
  • inp (Tensor): Current input \(\mathbf{x}(t)\),

    shape \((N, input\_size)\) or \((input\_size)\).

  • state (Tuple[Tensor, Tensor], optional): Previous hidden and cell states

    (\(\mathbf{h}(t-1)\), \(\mathbf{c}(t-1)\)), each of shape \((N, hidden\_size)\) or \((hidden\_size)\). Defaults to zeros.

Outputs:
  • new_state (Tensor): Updated hidden state \(\mathbf{h}(t)\),

    shape \((N, hidden\_size)\) or \((hidden\_size)\).

  • new_cstate (Tensor): Updated cell state \(\mathbf{c}(t)\),

    shape \((N, hidden\_size)\) or \((hidden\_size)\).

weight_ih

Input-to-hidden weights, shape (4*hidden_size, input_size), split into i, f, c, o gates along dim=0.

Type:

Tensor

weight_hh

Hidden-to-hidden weights, shape (4*hidden_size, hidden_size), split into i, f, c, o gates.

Type:

Tensor

weight_mh

Working-memory weights, shape (3*hidden_size, hidden_size), split into i, f, o contributions.

Type:

Tensor

bias_ih

Input biases, shape (4*hidden_size,).

Type:

Tensor

bias_hh

Hidden biases, shape (4*hidden_size,).

Type:

Tensor

bias_mh

Memory biases, shape (3*hidden_size,).

Type:

Tensor

Examples::
>>> cell = WMCLSTMCell(input_size=8, hidden_size=16)
>>> seq = torch.randn(12, 4, 8)    # seq length 12, batch size 4
>>> h = torch.zeros(4, 16)         # initial hidden state
>>> c = torch.zeros(4, 16)         # initial cell state
>>> outputs = []
>>> for t in range(12):
...     h, c = cell(seq[t], (h, c))
...     outputs.append(h)
__init__(input_size, hidden_size, bias=True, recurrent_bias=True, memory_bias=True, kernel_init=<function xavier_uniform_>, recurrent_kernel_init=<function xavier_uniform_>, memory_kernel_init=<function xavier_uniform_>, bias_init=<function zeros_>, recurrent_bias_init=<function zeros_>, memory_bias_init=<function zeros_>, device=None, dtype=None)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Parameters:

  • input_size (int)

  • hidden_size (int)

  • bias (bool)

  • recurrent_bias (bool)

  • memory_bias (bool)

  • kernel_init (Callable)

  • recurrent_kernel_init (Callable)

  • memory_kernel_init (Callable)

  • bias_init (Callable)

  • recurrent_bias_init (Callable)

  • memory_bias_init (Callable)

  • device (device | None)

  • dtype (dtype | None)

Methods

__init__(input_size, hidden_size[, bias, ...])

Initialize internal Module state, shared by both nn.Module and ScriptModule.

add_module(name, module)

Add a child module to the current module.

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

buffers([recurse])

Return an iterator over module buffers.

children()

Return an iterator over immediate children modules.

compile(*args, **kwargs)

Compile this Module's forward using torch.compile().

cpu()

Move all model parameters and buffers to the CPU.

cuda([device])

Move all model parameters and buffers to the GPU.

double()

Casts all floating point parameters and buffers to double datatype.

eval()

Set the module in evaluation mode.

extra_repr()

Return the extra representation of the module.

float()

Casts all floating point parameters and buffers to float datatype.

forward(inp[, state])

Run one step of the recurrent cell.

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

get_extra_state()

Return any extra state to include in the module's state_dict.

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

half()

Casts all floating point parameters and buffers to half datatype.

init_weights()

ipu([device])

Move all model parameters and buffers to the IPU.

load_state_dict(state_dict[, strict, assign])

Copy parameters and buffers from state_dict into this module and its descendants.

modules()

Return an iterator over all modules in the network.

mtia([device])

Move all model parameters and buffers to the MTIA.

named_buffers([prefix, recurse, ...])

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

named_modules([memo, prefix, remove_duplicate])

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

named_parameters([prefix, recurse, ...])

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

parameters([recurse])

Return an iterator over module parameters.

register_backward_hook(hook)

Register a backward hook on the module.

register_buffer(name, tensor[, persistent])

Add a buffer to the module.

register_forward_hook(hook, *[, prepend, ...])

Register a forward hook on the module.

register_forward_pre_hook(hook, *[, ...])

Register a forward pre-hook on the module.

register_full_backward_hook(hook[, prepend])

Register a backward hook on the module.

register_full_backward_pre_hook(hook[, prepend])

Register a backward pre-hook on the module.

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module's load_state_dict() is called.

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module's load_state_dict() is called.

register_module(name, module)

Alias for add_module().

register_parameter(name, param)

Add a parameter to the module.

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

requires_grad_([requires_grad])

Change if autograd should record operations on parameters in this module.

set_extra_state(state)

Set extra state contained in the loaded state_dict.

set_submodule(target, module[, strict])

Set the submodule given by target if it exists, otherwise throw an error.

share_memory()

See torch.Tensor.share_memory_().

state_dict(*args[, destination, prefix, ...])

Return a dictionary containing references to the whole state of the module.

to(*args, **kwargs)

Move and/or cast the parameters and buffers.

to_empty(*, device[, recurse])

Move the parameters and buffers to the specified device without copying storage.

train([mode])

Set the module in training mode.

type(dst_type)

Casts all parameters and buffers to dst_type.

uses_double_state()

Return True if forward returns (h, c), else just h.

xpu([device])

Move all model parameters and buffers to the XPU.

zero_grad([set_to_none])

Reset gradients of all model parameters.

Attributes

T_destination

call_super_init

dump_patches

weight_ih

weight_hh

weight_mh

bias_ih

bias_hh

bias_mh

input_size

hidden_size

bias

recurrent_bias

training