[ENH] TabPFN-TS foundation model · sktime/sktime#8664 | Good First Issue

(12 留言) (0 反應) (1 負責人)Python (1,192 fork)batch import

enhancementgood first issueinterfacing algorithmsmodule:forecasting

倉庫指標

Star: (7,162 star)
PR 合併指標: (平均合併 26天 10小時) (30 天內合併 86 個 PR)

描述

We should interface the TabPFN-TS model https://github.com/PriorLabs/tabpfn-time-series

Estimator interfacing guide: https://www.sktime.net/en/latest/developer_guide/add_estimators.html

To check:

license constraints and implications. For copied code, the custom license of Prior Labs needs to be respected.
I presume we need to use _multiton to avoid memory explosion when broadcasting. You can search for it or look at ChronosForecaster

Optional: add a second version: a delegator that combines make_reduction with vanilla TabPFN - TabPFN-TS is exactly that, internally.

貢獻者指南

研究方向: 審查 TabPFN TS 倉庫（https://github.com/PriorLabs/tabpfn time series）的授權限制和介面相容性。參考 sktime 開發者指南中關於新增估計器的部分（https://www.sktime.net/en/latest/developer guide/add estimators.html），並查閱 ChronosForecaster 的實作，了解如何使用 ` multiton` 來避免記憶體問題。檢查是否需要結合 `make reduction` 與原始 TabPFN 的委託器作為可選的第二個版本。
技術棧: python
領域: machine learning
議題類型: 功能
難度: 3
預計時間: 半天
活動狀態: 活躍
清晰度: 大致清晰
前置要求: GitPython
新手友善度: 70