WenjieDu/PyPOTS

A Python toolkit/library for reality-centric machine/deep learning & data mining on partially-observed time series, with 50+ SOTA neural network models for scientific analysis tasks (imputation, classification, clustering, forecasting, anomaly detection, cleaning) on incomplete industrial irregularly-sampled multivariate TS with NaN missing values

What it solves

PyPOTS 解決了分析包含缺失值的真實世界時間序列資料（部分觀測時間序列，簡稱 POTS）的挑戰。感測器故障或通訊錯誤導致的缺失資料常常阻礙進階資料分析與機器學習，而迄今為止，市場上缺乏專門且統一的工具套件來滿足這類需求。

How it works

PyPOTS 提供一套完整的 Python 工具箱，整合了大量經典與最先進的機器學習演算法，專門針對具有缺失值的多變量時間序列進行調整。它提供統一的 API 與詳細的文件說明，簡化這些模型的實作。對於原本未針對 POTS 設計的模型，函式庫會套用特定的嵌入策略與訓練方法（例如 ORT+MIT），使其能夠處理缺失資料。

Who it’s for

此工具箱適合需要處理時間序列缺失問題的研究人員與工程師，讓他們不必在繁瑣的資料前處理或手動實作演算法上花費過多時間。

Highlights

  • Diverse Task Support: Supports imputation, forecasting, classification, clustering, and anomaly detection.
  • Extensive Algorithm Library: Includes a vast range of models from naive methods (mean/median) to advanced Neural Networks, Time-Series Foundation Models (TSFM), and Large Language Models (LLM) like GPT4TS.
  • Hyperparameter Optimization: Integrated support for Optuna and Microsoft NNI for tuning neural network models.
  • Ecosystem Integration: Works alongside TSDB (for easy dataset loading) and PyGrinder (for simulating missing data patterns like MCAR, MAR, and MNAR).