sktime/sktime

[ENH] `StackingForecaster` regressor should be able to use `X` as well

Open

#9,264 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:forecasting

描述

Discussed in https://github.com/sktime/sktime/discussions/9251

Originally posted by RobKuebler January 16, 2026 Hi!

In the StackingForecaster, we train a regressor on the outputs of n forecasters. I.e., this regressor uses n features to make a prediction. But since we pass X to StackingForecaster anyway, why not let the regressor see the X as well? At least via a keyword like use_X or something better.

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。
half day
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
PythonsktimeStackingForecaster
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。
85
研究方向
Examine the implementation of StackingForecaster in the sktime codebase, likely in `sktime/forecasting/compose/ stack.py`. Understand how the regressor is trained and predicted on the forecasts. The feature would involve adding a parameter `use X` (or similar) and passing the exogenous variables `X` to the regressor's `fit` and `predict` methods. Review the linked discussion for additional considerations.