Explainability

Functionality to help with both global and local explainability.

---

source

get_attribution_map

 get_attribution_map (model, modules, x, y=None, detach=True, cpu=False,
                      apply_relu=True)

---

source

get_acts_and_grads

 get_acts_and_grads (model, modules, x, y=None, detach=True, cpu=False)

Returns activations and gradients for given modules in a model and a single input or a batch. Gradients require y value(s). If they are not provided, it will use the predictions.