scikit-learn/scikit-learn

[WIP] Online Inference for BayesianGaussianMixture

Open

#8,714 opened on 2017年4月6日

GitHub で見る
 (5 comments) (0 reactions) (0 assignees)Python (66,084 stars) (27,020 forks)batch import
Enhancementhelp wantedmodule:mixture

説明

It would be great if someone implemented partial_fit for BayesianGaussianMixture using mean field stochastic variational inference. Fitting with EM is way too slow to use for online clustering on large datasets, and right now the only online clustering method in sklearn is MiniBatchKMeans, which doesn't handle noisy data very well. Alternatively, a parallel Gibbs sampler would work too but is likely a lot more work.

コントリビューターガイド

技術スタック
pythonnumpy
領域
machine learningdata
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。
5
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。
over 1 week
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。
clear
前提条件
Pythonscikit learnBayesian statisticsvariational inference
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。
10
調査方針
This issue requests adding online inference (partial fit) to BayesianGaussianMixture using mean field stochastic variational inference (SVI) as described in Hoffman et al. (2013). The current implementation uses EM and is located in sklearn/mixture/. To implement this, one should study the SVI algorithm for Gaussian mixtures and adapt the existing code. Alternatively, a parallel Gibbs sampler could be considered but is more complex. Refer to MiniBatchKMeans for an example of online clustering in scikit learn.