SpanRepresentationScorerLayer<T>

Layers · AiDotNet.NeuralNetworks.Layers

Scores every span (start, end) of a sentence against every entity category from a span representation: the endpoint representation of PURE or the max-pooled representation of SpERT, each with a learned width embedding.

Inheritance: LayerBase<T> → SpanRepresentationScorerLayer<T>

Implements: IShapeContract

How It Works

Shapes: input [B, S, D] (or [S, D]) token states; output [B, S * S, C], the flattened span grid with index start * S + end, the layout of BiaffineSpanScorerLayer. Spans with end < start are scored too (as the empty width) and are expected to be masked by the targets.

Widths above maxWidth share the last width embedding. In SpERT’s representation the first token’s state stands in for the sentence’s [CLS] vector, since the input carries no separate classification token.

Constructors

Constructor Summary
SpanRepresentationScorerLayer(SpanRepresentationKind kind, int inputDim, int widthDim, int maxWidth, int hiddenDim, int numCategories) Creates the scorer.

Properties

Property Summary
bool SupportsTraining { get; }

Methods

Method Summary
IReadOnlyList<ILayer<T>> GetSubLayers()
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank)
void ResetState()