PyPOTS
Python toolkit for machine learning on partially-observed time series with missing values.
Over 50 SOTA neural network models for analysis tasks on irregularly-sampled multivariate time series.
PyPOTS is a dedicated Python toolbox designed to make machine learning on partially-observed time series easy rather than tedious. It suits data scientists and engineers who need to handle real-world datasets plagued by missing sensor values, communication errors, and unexpected malfunctions.
What it is
The library provides unified APIs for 50+ state-of-the-art models supporting imputation, forecasting, classification, clustering, and anomaly detection tasks. It adapts models that do not natively handle missing data by applying specific embedding and training approaches like ORT and MIT.
Who it's for
It suits machine learning engineers and academic researchers working with multivariate industrial time series containing NaN values. It does not suit practitioners looking for general-purpose tabular imputation or simple statistical interpolation without deep learning.
Key features
Unified Model APIs
Access over 50 state-of-the-art neural network algorithms through consistent and standardized interfaces.
Multiple Core Tasks
Perform imputation, forecasting, classification, clustering, and anomaly detection using a single library.
Adapted Transformer Models
Run models like Transformer, Informer, and iTransformer directly on incomplete time series via custom embedding strategies.
Hyperparameter Optimization
Tune model hyperparameters smoothly utilizing integrated Optuna and Microsoft NNI support.
Up and running in 3 steps
Our honest verdict
What's great
- Fills a major tooling gap for time series with missing values
- Integrates recent state-of-the-art models from top machine learning venues
- Provides runnable code examples for each algorithm and task
- Includes representation learning capabilities via TS2Vec
Where it falls short
- Requires deep learning expertise and familiarity with Python data stacks
- Training numerous deep neural network models demands substantial compute resources
- Handling large multivariate datasets with missing values can be memory-intensive
Pricing & license
The toolkit is free and open source under its specified licence, with the only cost being the compute infrastructure required to train the deep learning models yourself.
Open remote AI-training roles
873 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
What types of tasks does PyPOTS support?
It supports imputation, forecasting, classification, clustering, and anomaly detection on multivariate time series.
Can standard transformer models handle missing values in PyPOTS?
Yes, PyPOTS applies specific embedding and training approaches to adapt models like iTransformer and Informer for incomplete data.
