Fix RISE algorithm for explain function

[wonjuleee]

Summary

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added the description of my changes into CHANGELOG.​
• I have updated the documentation accordingly

License

• I submit my code changes under the same MIT License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2023 Intel Corporation
#
# SPDX-License-Identifier: MIT

[codecov]

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (784e039) 80.54% compared to head (690b6b2) 80.61%.
Report is 1 commits behind head on develop.

Files	Patch %	Lines
src/datumaro/components/algorithms/rise.py	92.15%	2 Missing and 2 partials ⚠️
.../openvino_plugin/samples/googlenet-v4-tf_interp.py	71.42%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1263      +/-   ##
===========================================
+ Coverage    80.54%   80.61%   +0.06%     
===========================================
  Files          271      271              
  Lines        30438    30382      -56     
  Branches      5930     5909      -21     
===========================================
- Hits         24517    24491      -26     
+ Misses        4532     4507      -25     
+ Partials      1389     1384       -5     

Flag	Coverage Δ
ubuntu-20.04_Python-3.8	80.59% <92.68%> (+0.06%)	⬆️
windows-2022_Python-3.8	80.57% <92.68%> (+0.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sooahleex]

Looks good to me. I left some minor comments.

[sooahleex]

Suggested change

	normalized_saliency = np.empty_like(saliency)
	for idx, sal in enumerate(saliency):
	normalized_saliency[idx, ...] = (sal - np.min(sal)) / (np.max(sal) - np.min(sal))
	return normalized_saliency
	return (saliency - np.min(saliency)) / (np.max(saliency) - np.min(saliency))

I think this could be replaced like this through vectorized operation.

[wonjuleee]

thank you for the feedback, but the original code is to normalize each saliency map per class_idx, while your suggestion is to normalize each saliency map from global max and min. So I expect the results will be different.

[sooahleex]

Suggested change

	logit = pred.get(self.LOGIT_KEY)
	if logit is None:
	raise DatumaroError(f'"{self.LOGIT_KEY}" key should exist in the model prediction.')
	logit = pred.get(self.LOGIT_KEY, [])
	if not logit:
	raise DatumaroError(f'"{self.LOGIT_KEY}" key should exist in the model prediction.')

It seems better to use default values for safer.

[wonjuleee]

thank you. let me update this later.

[sooahleex]

Suggested change

	feature_vector = pred.get(self.FEAT_KEY)
	if feature_vector is None:
	raise DatumaroError(f'"{self.FEAT_KEY}" key should exist in the model prediction.')
	feature_vector = pred.get(self.FEAT_KEY, [])
	if not feature_vector:
	raise DatumaroError(f'"{self.FEAT_KEY}" key should exist in the model prediction.')