ICreditAssignmentContext<T>

Interfaces · AiDotNet.Interfaces

The information an ICreditRule receives for a single training step: the ordered trainable layers (with output shapes and current weights) and the network output error signal.

Properties

Property Summary
INumericOperations<T> NumOps { get; } Numeric operations for the type T.
IReadOnlyList<ICreditLayer<T>> Layers { get; } The trainable layers in forward (input → output) order.
Random Random { get; } A deterministic random source the rule may use to build fixed feedback matrices, for reproducibility.
Tensor<T> OutputError { get; } The output error signal prediction − target, shaped [batchSize, outputFeatures].
Tensor<T> Target { get; } The training target aligned to the prediction shape [batchSize, outputFeatures] — a one-hot matrix for classification, or the raw regression target.
int BatchSize { get; } The number of samples in this batch.