fix #137 and #181 (faulty ROC AUC scores)

Open robinruff opened this issue 2 years ago • 3 comments

Matbench Pull Request Template

Core code/data/docs changes

Brief description of changes

Fix for issues #137 and #181 (Faulty ROC AUC scores, caused by discretized probabilities). Removal of the homogenize_clf_array function in data_ops.py as it is no longer used.

Tests

Tests have been changed accordingly. Test for homogenize_clf_array was deleted, because the function does not exist anymore. Test for score_array has been corrected.

Closed issues or PRs

Issues: #137, #181

Mar 14 '23 21:03 robinruff

Great, thanks for the PR! I will review and merge soon

Mar 17 '23 22:03 ardunn

As a point of discussion, are there thoughts on using F1 as a replacement for ROCAUC for the leaderboards? Since many of the algorithms only have discretized probability $\in {0,1}$

Apr 07 '23 04:04 ardunn

@hrushikesh-s @computron thoughts on this?

Apr 07 '23 20:04 ardunn