scikit-learn/scikit-learn

parallel votingClassifier prediction method needed

Open

#7,755 opened on 2016年10月26日

GitHub で見る
 (3 comments) (0 reactions) (0 assignees)Python (66,084 stars) (27,020 forks)batch import
help wantedmodule:ensemble

説明

Hi there: I am using a really big jungle of single classifiers for an ensemble voting, something like:

emclf = VotingClassifier(estimators=est_list,n_jobs=-1) the est_list has about 300 classifiers. ---->that's all good and thank you folks.

However, when I am done training and doing predictions, like :

prd = emclf.predict(test_x)

the prediction method is not conducted in a parallel way, which became a bottleneck for the performance.

Is that possible to make this predict step parallel? thanks

コントリビューターガイド

技術スタック
python
領域
machine learning
Issue 種別
feature
難度新しい貢献者にとっての実装難度です。1 は非常に小さな変更、5 は専門的な作業です。
3
推定時間経験ある貢献者が調査、実装、テスト、pull request 準備にかけるおおよその時間範囲です。
1-2 days
活動状況Issue が今どれだけ取り組みやすいかを示します。新しい、活発、古い、ブロック中、またはメンテナー入力待ちなどです。
stale
明確さ期待される変更、受け入れ条件、次の手順がどれだけ明確かを示します。
mostly clear
前提条件
familiarity with scikit learn's VotingClassifierunderstanding of joblib parallelismexperience with scikit learn's estimator API
初心者向け度初回貢献者にどれだけ取り組みやすいかを 1-100 で推定したスコアです。
35
調査方針
The issue requests parallelizing the predict method of VotingClassifier, similar to how n jobs= 1 works for fit. Investigate the current VotingClassifier implementation in sklearn/ensemble/ voting.py, where the predict method iterates over estimators sequentially. The existing fit method uses joblib.Parallel with n jobs. The research should focus on adapting the predict method to use joblib.Parallel for collecting predictions from each estimator, then aggregating them. The merge strategy is already in place; only the parallel loop needs to be added. The issue is clear but requires understanding of scikit learn's parallelization patterns.
parallel votingClassifier prediction method needed · scikit-learn/scikit-learn#7755 | Good First Issue