8000 Inconsistent init_estimator for MultinomialDeviance · Issue #13574 · scikit-learn/scikit-learn · GitHub
[go: up one dir, main page]
Skip to content

Inconsistent init_estimator for MultinomialDeviance #13574

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
yanyang82 opened this issue Apr 4, 2019 · 1 comment
Closed

Inconsistent init_estimator for MultinomialDeviance #13574

yanyang82 opened this issue Apr 4, 2019 · 1 comment

Comments

@yanyang82
Copy link

In gradient_boosting.py, the MultinomialDeviance defines its init_estimator to be an instance of PriorProbabilityEstimator, which predicts the sample proportions for each class label as the prior. This is inconsistent with _score_to_proba which maps the score (log likelihood) to actual probabilities

The correct init estimator should use the log of the sample proportions for each class label, which then recovers to probability when substituting into _score_to_proba.

Both BinomialDeviance and ExponentialLoss are consistent in that regard in that their init_estimator predicts log likelihood and half log likelihood respectively, which are the target value each regression tree is predicting.
@yanyang82
Copy link
Author

Just realized this is already fixed in #12983, closing the issue

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@yanyang82
0