struct

`SushiAI::NN::Linear`

Represents an affine transformation: Y = X * W^T + b.

Public attributes

int64_t in_features = 0

Input features per sample.

int64_t out_features = 0

Output features per sample.

bool use_bias = true

Whether to add a learnable bias.

Parameter weight {}

The [out_features, in_features] weight.

Parameter bias {}

The [out_features] bias; undeclared when use_bias is false.

static constexpr auto fields()

This layer's parameters, for the generic walk.

Graph::ValueId forward(const Graph::GraphBuilder &builder, Graph::ValueId x)

Traces the layer, declaring its parameters on first call.

Parameters

Returns

The [N, out_features] result's id.

Exceptions

Shape output_shape(const Shape &input) const

Returns the per-sample shape forward() produces, [out_features].

Parameters