peal.entrypoints.run_explainer¶
Explain a trained predictor with a stand-alone explainer run.
Builds the explainer that explainer_type in the config selects through
peal.explainers.explainer_factory.get_explainer (chiefly the
counterfactual explainer in peal/explainers/counterfactual_explainer.py),
runs it over the dataset it is configured with, has the configured teacher
annotate the explanations and renders the interpretations.
Invocation:
python run_explainer.py --config <explainer.yaml> \
[--oracle_path <model.cpl>] [--confounder_oracle_path <model.cpl>]
Explainer configs live in configs/*/explainers. The explanations,
collages and interpretation figures are written into the config’s
explanations_dir. The two oracle paths are forwarded to
explainer.run and the models are loaded afterwards, but the evaluation
against them that the trailing branches announce was never implemented: the
loaded oracles are unused.
Functions
Build the explainer from |