attribench.functional.metrics.irof

attribench.functional.metrics.irof(model, attributions_dataset, batch_size, maskers, activation_fns='linear', mode='morf', start=0.0, stop=1.0, num_steps=100, device=device(type='cpu'))

Computes the IROF metric for a given AttributionsDataset and model.

IROF starts segmenting the input image using SLIC. Then, it iteratively masks out the top (Most Relevant First, or MoRF) or bottom (Least Relevant First, or LeRF) segments and computes the confidence of the model on the masked samples. The relevance of a segment is computed as the average relevance of the features in the segment.

This results in a curve of confidence vs. number of segments masked. The area under (or equivalently over) this curve is the IROF metric.

start, stop, and num_steps are used to determine the range of segments to mask. The range is determined by start and stop as a percentage of the total number of segments. num_steps is the number of steps to take between start and stop.

The IROF metric is computed for each masker in maskers and for each activation function in activation_fns.

Parameters:

modelnn.Module

Model to compute IROF on.

attributions_datasetAttributionsDataset

Dataset of attributions to compute IROF on.

batch_sizeint

Batch size to use when computing model predictions on masked samples.

maskersMapping[str, ImageMasker]

Dictionary of maskers to use for masking samples.

activation_fnsUnion[List[str], str], optional

List of activation functions to use when computing model predictions on masked samples. If a single string is passed, it is converted to a single-element list. Default: “linear”

modestr, optional

Mode to use when masking samples. Must be “morf” or “lerf”. Default: “morf”

startfloat, optional

Relative start of the range of segments to mask. Must be between 0 and 1. Default: 0.0

stopfloat, optional

Relative stop of the range of segments to mask. Must be between 0 and 1. Default: 1.0

num_stepsint, optional

Number of steps to take between start and stop. Default: 100