percent_better_than_baseline fails when baseline is 0

Repro: dataset X = pd.read_csv('/Users/becca.mcbrayer/Desktop/text_datasets/wine_modified.csv')[:100] y = X.pop('points') X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42) automl_no_text = AutoMLSearch(problem_type='binary', objective='precision') automl_no_text.search(X_train, y_train)

Output:

[eccabay]

I see, I must have missed that!

[kmax12]

that’s true, but if you think about it has a percent improvement or decrease inf would make sense too. that being said, it’s a bit weird in cases where the bounds on the metric are something like (-1, 1) since a zero score would give you inf, but a slight deviation wouldnt.

so, i guess i convinced myself nan is right