Dumping robustness check results

We provide two standard ways for dumping the robustness check results:

robustcheck.utils.save_robustness_stats_artifacts(robustness_check, run_output_folder)

Saves robustness check artifacts containings metrics and histograms of queries and perturbartion distances on the local file system.

Parameters:

• robustness_check – RobustnessCheck containing the model and dataset to be benchmarked. This requires its run_robustness_check() method to have been executed such that we have metrics to extract from it.

• run_output_folder – A string representing where to save the arising artifacts.

robustcheck.utils.generate_mlflow_logs(robustness_check, run_name, experiment_name='default', tracking_uri='mlruns')

Generates robustness check logs on mlflow.

Parameters:

• robustness_check – RobustnessCheck containing the model and dataset to be benchmarked. This requires its run_robustness_check() method to have been executed such that we have metrics to extract from it.

• run_name – A string representing the run name under which the mlflow artifacts and metrics will be logged.

• experiment_name – A string representing the experiment name under which the mlflow artifacts and metrics will be logged.

• tracking_uri – A string representing the path where the mlflow artifacts and metrics will be stored.