sktime/sktime

[ENH] Add native probabilistic prediction methods to `Chronos2Forecaster`

Open

#10,227 建立於 2026年5月11日

在 GitHub 查看
 (3 留言) (0 反應) (1 負責人)Python (7,162 star) (1,192 fork)batch import
enhancementgood first issue

描述

Is your feature request related to a problem? Please describe.

If I'm reading the original code and paper correctly, Chronos2Forecaster the model supports probabilistic predictions natively, at least predict_quantiles. This should be added to the sktime wrapper. Ideally, we also add predict_proba but this would require more work to find a proper way to wire the forecaster through skpro.

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。
blocked
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
Pythonsktime librarytime series forecasting
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。
35
研究方向
Examine the existing Chronos2Forecaster implementation in sktime (likely in sktime/forecasting/chronos) to understand the current wrapper. Check the original Chronos2 model's predict quantiles and predict proba methods. Review the skpro library to understand how probabilistic forecasts are wired. The issue comments may contain further guidance from maintainers. Start by implementing predict quantiles, then explore predict proba.