---
id: pypots
version: "1.5"
license: Copyright (c) 2023-present, Wenjie Du All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are… (full text in the JSON record)
license_treatment: permissive
maintenance: active
---

# pypots — A Python Toolbox for Machine Learning on Partially-Observed Time Series

License: permissive · Maintenance: active · Downloads: 123.0K/mo

## What it is and what it does

PyPOTS is a machine learning toolkit designed specifically for time series data with missing values—a common real-world problem caused by sensor failures, communication errors, or data collection issues. It fills a gap by providing a unified interface to classical and state-of-the-art algorithms for handling partially-observed time series, rather than requiring researchers to build custom solutions or adapt general-purpose libraries.

The package supports five core tasks: imputation (filling missing values), forecasting, classification, clustering, and anomaly detection. It includes neural network models, transformers adapted for incomplete data, and LLM-based approaches. All models share consistent APIs and come with documentation and examples. The toolkit depends on torch, transformers, scikit-learn, and other standard data science libraries, making it suitable for researchers and engineers working with real-world time series that contain gaps.

Use it for:

- Impute missing sensor readings in IoT or industrial monitoring data before downstream analysis
- Forecast equipment failures or system behavior when historical data has communication gaps
- Classify time series patterns (e.g., patient conditions) despite incomplete medical records
- Detect anomalies in multivariate time series with irregular sampling or missing observations
- Cluster similar time series trajectories when some features are not always recorded

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

PyPOTS provides machine learning algorithms for time series with missing values, supporting imputation, classification, clustering, forecasting, and anomaly detection on partially-observed multivariate data.

Yes, if you work with incomplete time series data and need a unified toolkit with multiple algorithms. The package is actively maintained, BSD-licensed, and has no known vulnerabilities. The 16 runtime dependencies (especially torch and transformers) add significant install size and complexity; evaluate whether you need the full suite or can use a lighter alternative if you only need imputation or forecasting.

## Install

pip install pypots
uv add pypots
poetry add pypots

## Installing pypots

Before you install: Low friction installation as a pure Python wheel. Active maintenance with a recent release (101 days ago) and steady repository activity. Depends on 16 runtime packages including torch, transformers, and scikit-learn, which are themselves well-maintained but add substantial disk footprint.

License in practice: BSD 3-Clause license (permissive). You may use, modify, and distribute the package freely in commercial and private projects provided you retain the copyright notice and disclaimer.

Quickstart:

pip install pypots

import pypots
from pypots.imputation import SAITS

# Initialize and use an imputation model
model = SAITS(n_steps=24, n_features=10)
model.fit(train_data)
imputed = model.predict(test_data)

Requires torch and transformers, which have their own system dependencies (CUDA optional but recommended for GPU acceleration). Python 3.8 or later.

Verify before relying:

- Whether all listed algorithms (HELIX, MixLinear, SegRNN, etc.) are production-ready or still experimental
- Performance benchmarks comparing PyPOTS models to standalone implementations
- Memory requirements for typical multivariate time series datasets
- Whether hyperparameter optimization via NNI is included in the base install or requires separate setup

## Package facts

- License: Copyright (c) 2023-present, Wenjie Du All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are… (full text in the JSON record) (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 123.0K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

time series missing data imputation, incomplete time series machine learning, partially observed time series, time series forecasting with gaps, time series anomaly detection missing values, multivariate time series classification, time series clustering incomplete data, time-series, missing-data, neural-networks

[View on SkillFed](https://skillfed.io/packages/pypots) · [View on PyPI](https://pypi.org/project/pypots/)