Distributed API Reference

This page contains the reference for the distributed API of attribench. For simple, small-scale experiments, the attribench functional API is recommended. For large-scale experiments, the distributed API is recommended, as it can leverage multiple GPUs.

General

attribench.distributed.SelectSamples(...[, ...])	Select correctly classified samples from a dataset and write them to a HDF5 file.
attribench.distributed.TrainAdversarialPatches(...)	Train adversarial patches for a given model and dataset and save them to disk.
attribench.distributed.ComputeAttributions(...)	Compute attributions for a dataset using multiple processes.

Metrics

attribench.distributed.Metric(model_factory, ...)	Abstract base class for metrics that are computed using multiple processes.
attribench.distributed.metrics.Deletion(...)	Compute the Deletion metric for a given AttributionsDataset and model using multiple processes.
attribench.distributed.metrics.Insertion(...)	Compute the Insertion metric for a given AttributionsDataset and model using multiple processes.
attribench.distributed.metrics.ImpactCoverage(...)	Computes the Impact Coverage metric for a given dataset, model, and set of attribution methods, using multiple processes.
attribench.distributed.metrics.Irof(...[, ...])	Compute the IROF metric for a given AttributionsDataset and model using multiple processes.
attribench.distributed.metrics.Infidelity(...)	Computes the Infidelity metric for a given AttributionsDataset and model using multiple processes.
attribench.distributed.metrics.MaxSensitivity(...)	Compute the Max-Sensitivity metric for a given Dataset and attribution methods using multiple processes.
attribench.distributed.metrics.MinimalSubset(...)	Computes Minimal Subset Deletion or Insertion for a given AttributionsDataset and model using multiple processes.
attribench.distributed.metrics.SensitivityN(...)	Compute the Sensitivity-n metric for a given AttributionsDataset and model using multiple processes.