Module nn

Module nn 

Source
Expand description

A small neural-network library based on Torch.

This library tries to stay as close as possible to the original Python and C++ implementations.

Re-exports§

pub use init::f_init;
pub use init::init;
pub use init::Init;

Modules§

init
Variable initialization.

Structs§

Adam
Parameters for the Adam optimizer.
AdamW
Parameters for the AdamW optimizer.
BatchNorm
A batch-normalization layer.
BatchNormConfig
Batch-normalization config.
Conv
A N-dimensional convolution layer.
ConvConfigND
Generic convolution config.
ConvTransposeConfigND
A generic transposed convolution configuration.
ConvTransposeND
A generic transposed convolution layer.
Embedding
An embedding layer.
EmbeddingConfig
Configuration option for an embedding layer.
Func
A layer defined by a simple closure.
FuncT
A layer defined by a closure with an additional training parameter.
GRU
A Gated Recurrent Unit (GRU) layer.
GRUState
A GRU state, this contains a single tensor.
GroupNorm
A group-normalization layer.
GroupNormConfig
Group-normalization config.
Id
An identity layer. This just propagates its tensor input as output.
LSTM
A Long Short-Term Memory (LSTM) layer.
LSTMState
The state for a LSTM network, this contains two tensors.
LayerNorm
A layer-normalization layer.
LayerNormConfig
Layer-normalization config.
Linear
A linear fully-connected layer.
LinearConfig
Configuration for a linear layer.
Optimizer
An optimizer to run gradient descent.
Path
A variable store with an associated path for variables naming.
RNNConfig
Configuration for the GRU and LSTM layers.
RmsProp
Parameters for the RmsProp optimizer.
Sequential
A sequential layer combining multiple other layers.
SequentialT
A sequential layer combining new layers with support for a training mode.
Sgd
Parameters for the SGD optimizer.
VarStore
A VarStore is used to store variables used by one or multiple layers. It specifies a single device where all variables are stored.
Variables

Enums§

PaddingMode
How padding is performed by convolution operations on the edge of the input tensor.

Traits§

Module
The simplest module trait, defining a forward function.
ModuleT
Module trait with an additional train parameter.
OptimizerConfig
Optimizer configurations. These configs can be used to build optimizer.
RNN
Trait for Recurrent Neural Networks.

Functions§

adam
Creates the configuration for the Adam optimizer.
adamw
Creates the configuration for the AdamW optimizer.
batch_norm1d
Applies Batch Normalization over a three dimension input.
batch_norm2d
Applies Batch Normalization over a four dimension input.
batch_norm3d
Applies Batch Normalization over a five dimension input.
conv
Creates a new convolution layer for any number of dimensions.
conv1d
Creates a new one dimension convolution layer.
conv2d
Creates a new two dimension convolution layer.
conv3d
Creates a new three dimension convolution layer.
conv_transpose1d
Creates a one dimension transposed convolution layer.
conv_transpose2d
Creates a two dimension transposed convolution layer.
conv_transpose3d
Creates a three dimension transposed convolution layer.
embedding
func
func_t
group_norm
gru
Creates a new GRU layer.
layer_norm
linear
Creates a new linear layer.
lstm
Creates a LSTM layer.
no_bias
The default convolution config without bias.
rms_prop
Creates the configuration for the RmsProp optimizer.
seq
Creates a new empty sequential layer.
seq_t
Creates a new empty sequential layer.
sgd
Creates the configuration for a Stochastic Gradient Descent (SGD) optimizer.

Type Aliases§

Conv1D
One dimension convolution layer.
Conv2D
Two dimensions convolution layer.
Conv3D
Three dimensions convolution layer.
ConvConfig
Convolution config using the same parameters on all dimensions.
ConvTranspose1D
A one dimension transposed convolution layer.
ConvTranspose2D
A two dimension transposed convolution layer.
ConvTranspose3D
A three dimension transposed convolution layer.
ConvTransposeConfig
A transposed convolution configuration using the same values on each dimension.