Why am I getting different learning curve for SVC in using sklearn and sklearnex about scikit-learn-intelex HOT 5 CLOSED

sklearn and sklearnex implementations of SVM are expected to show different results.
What score is used as loss in these learning curves?

from scikit-learn-intelex.

In here I used f1 score as the matric, the label of the y-axis might be wrong. So which result should I trust,or both are correct?

from scikit-learn-intelex.

Formally, flattened learning curve for training/validation and higher f1-score for sklearnex is not an issue. It looks like sklearn has convergence issue for bigger number of samples.

from scikit-learn-intelex.

Score issue is not an issue in this case.

from scikit-learn-intelex.

Reference back to stackoverflow post - https://stackoverflow.com/questions/76842446/why-am-i-getting-different-learning-curve-for-svc-in-using-sklearn-and-sklearnex and answer from there

"I ran a somewhat similar experiment, and got consistent results between sklearn/sklearnex:

I was using SVR with 5-fold CV and a similarly-sized dataset to yours. I ran it under sklearn 1.3.0. and sklearnex 2023.2.1
"

from scikit-learn-intelex.