experiment_runner.main

Entry point for the xai4tsc experiment runner.

Run from the repository root as a module:

python -m experiment_runner.main --conf path/to/config.yaml

Orchestrates the full evaluation pipeline as a nested loop over datasets, models, explainers, and metrics. Results are saved as metrics.csv files at three granularity levels (global, per-dataset, per-model) under results_rel_path.

Attributes

logger
MASTER_CONFIG
parser

Functions

main(→ int)	Run a complete XAI evaluation experiment from a YAML configuration file.
initial_setup(→ tuple[dict, logging.Handler])	Set up experiment as described in the given configuration file.

Module Contents

experiment_runner.main.logger

experiment_runner.main.MASTER_CONFIG

experiment_runner.main.main(config_path: str, debug: bool = False) → int

Run a complete XAI evaluation experiment from a YAML configuration file.

Orchestrates the full pipeline as a nested loop over datasets → models → explainers → metrics. Results are saved as metrics.csv at three granularity levels (global, per-dataset, per-model) under results_rel_path.

Failures are isolated at the level they occur: a raising metric costs only that metric, a raising explainer only that explainer, and a raising model only that model, so the rest of the dataset’s results are still collected and written.

Returns:

Total number of failures across all four levels (datasets, models, explainers and metric evaluations). 0 means everything configured either completed or was explicitly skipped, so callers can treat a non-zero value as “this run needs attention” (the CLI turns it into a non-zero exit status).

Return type:

int

experiment_runner.main.initial_setup(config_path: str, debug: bool = False, issues: experiment_runner.config.ConfigIssues | None = None) → tuple[dict, logging.Handler]

Set up experiment as described in the given configuration file.

Parameters:

• config_path (str) – Path to the experiment configuration.

• debug (bool) – Enable debug-level logging.

• issues (ConfigIssues, optional) – Collector for config warnings found here. Deferring them is what gets them into experiment.log, which does not exist yet at this point.

Returns:

The extracted configuration and the run-level log handler, which the caller detaches when the run ends.

Return type:

tuple[dict, logging.Handler]

experiment_runner.main.parser