sktime/sktime

[ENH] Add caching to pipelines

Open

#9,263 建立於 2026年1月19日

在 GitHub 查看
 (3 留言) (0 反應) (0 負責人)Python (7,162 star) (1,192 fork)batch import
enhancementgood first issuemodule:base-frameworkmodule:classificationmodule:clusteringmodule:detectionmodule:forecastingmodule:transformations

描述

Discussed in https://github.com/sktime/sktime/discussions/9249

Originally posted by RobKuebler January 16, 2026 Hi!

scikit-learns's Pipeline has the keyword memory that lets you cache pipeline steps. This has the effect that hyperparameter tuning can be done much faster since very often only a few steps of the pipeline (often only the model) is refitted.

sktime would benefit from this as well.

貢獻者指南

技術棧
python
領域
machine learning
議題類型
feature
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
3
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
1-2 days
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
fresh
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
clear
前置要求
familiarity with sktime pipelinesunderstanding of sklearn's memory parameter
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
45
研究方向
Review the discussion at https://github.com/sktime/sktime/discussions/9249 for context. Examine sktime's pipeline implementation in the source code (likely `sktime/pipeline.py` or similar) to identify where the `memory` parameter from sklearn's Pipeline can be integrated. Look at how sklearn implements caching with joblib or similar, and propose a minimal implementation that adds a `memory` keyword to sktime's Pipeline class, allowing users to cache intermediate steps during hyperparameter tuning.