shap-result-analysis

shap-result-analysis runs a survival-model result pipeline around recommendations produced by shap-recommender. It supports exclusion, non-linearity and interaction variants, independent confirmation, comparator models, sequential ablation, calibration, proportional-hazards checks and subgroup reporting.

Installation

pip install shap-result-analysis

Input files

The pipeline expects a data directory containing feature_spec.json and the following tab-separated files:

• X_disc_model.tsv and y_disc.tsv
• X_conf_model.tsv and y_conf.tsv
• X_train_model.tsv and y_train.tsv
• X_test_model.tsv and y_test.tsv
• X_train_original.tsv and X_test_original.tsv

Each outcome file must contain event and time columns. By default the data directory is rebuttal_data, matching the source pipeline.

Command line

Run the full pipeline from the directory containing the data directory:

shap-result-analysis

Optional stages can be skipped:

shap-result-analysis --skip-validation --skip-comparators --skip-ablation

Library use

from result_analysis import main

main(
    run_validation=True,
    run_comparators=True,
    run_ablation=True,
    run_dose_response=True,
    run_margin=False,
)

License

MIT