Crate tsai_explain 

tsai_explain

Explainability tools for tsai-rs: activation/gradient capture, attribution maps.

This crate provides:

• Activation capture hooks
• Gradient capture hooks
• Attribution maps (CAM-like for CNNs, attention for transformers)

Structs

ActivationCapture

Captured activations from model layers.

AttributionMap

Attribution map over time series.

GradientCapture

Captured gradients from model layers.

IntegratedGradientsConfig

Configuration for Integrated Gradients.

Enums

AttentionAggregation

How to aggregate attention weights across layers.

AttributionMethod

Method for computing attribution maps.

BaselineType

Type of baseline to use for Integrated Gradients.

Functions

attention_attribution

Compute Attention-based attribution for transformer models.

grad_cam

Compute GradCAM attribution for CNN models.

input_gradient

Compute Input × Gradient attribution.

integrated_gradients

Compute Integrated Gradients attribution.

random_baseline

Create a random baseline for Integrated Gradients.

zero_baseline

Create a zero baseline for Integrated Gradients.