---
id: benchpots
version: "1"
license: Copyright (c) 2024-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
---

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

License: permissive · Maintenance: active · Downloads: 122.4K/mo

## What it is and what it does

BenchPOTS is a benchmarking toolkit for evaluating machine learning algorithms on partially-observed time series (POTS)—datasets with missing or irregularly sampled values. It provides unified preprocessing pipelines for standard POTS datasets and a suite of evaluation tasks to measure algorithm performance on problems like imputation, classification, clustering, and forecasting.

The package wraps datasets like PhysioNet2012 and applies configurable missing-data patterns (e.g., MCAR with specified rates) to create reproducible benchmarks. It depends on torch, scikit-learn, pandas, numpy, and h5py for core functionality, plus specialized time-series libraries (tsdb, pygrinder, nonlinear_benchmarks). It targets researchers and practitioners developing or comparing algorithms for incomplete time series.

Use it for:

- Evaluate imputation algorithms on standard POTS datasets with controlled missing-data rates.
- Benchmark time-series forecasting models on partially-observed data from medical or sensor domains.
- Compare classification and clustering performance across multiple incomplete time-series datasets.
- Preprocess and standardize POTS datasets for reproducible machine learning research.
- Validate new time-series algorithms against established benchmarks before publication.

## Worth the install?

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

BenchPOTS provides standardized preprocessing pipelines and evaluation tasks for benchmarking machine learning algorithms on partially-observed time series datasets.

Yes, if you are developing or comparing machine learning algorithms for partially-observed time series. The package offers low install friction, active maintenance, permissive licensing, and no known vulnerabilities. Install with caution if you have limited disk or memory—the 8 runtime dependencies (especially torch) are substantial. Not necessary for one-off time-series tasks; best suited for research and systematic benchmarking.

## Install

pip install benchpots
uv add benchpots
poetry add benchpots

## Installing benchpots

Before you install: Low install friction with a pure Python wheel. Active maintenance with a recent commit on 2026-08-03. Depends on established scientific libraries (numpy, pandas, scikit-learn, torch, h5py) plus specialized time-series packages (tsdb, pygrinder, nonlinear_benchmarks).

License in practice: BSD License (permissive) allows commercial and private use with minimal restrictions—only requires retaining copyright notice and disclaimer in distributions.

Quickstart:

pip install benchpots

import benchpots
benchpots.datasets.preprocess_physionet2012(subset="all", rate=0.1)

Requires Python >=3.8; torch and h5py may need system-level dependencies (CUDA, HDF5 libraries) depending on your environment.

Verify before relying:

- Whether all 8 runtime dependencies (h5py, numpy, pandas, scikit-learn, torch, tsdb, pygrinder, nonlinear_benchmarks) are required for basic usage or only for specific datasets/tasks.
- Whether the package includes pre-downloaded datasets or requires separate data acquisition.
- Performance characteristics and memory footprint when working with large-scale time series.

## Package facts

- License: Copyright (c) 2024-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: 122.4K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

time series benchmarking, partially observed time series, incomplete time series evaluation, missing data imputation benchmark, time series preprocessing pipeline, POTS dataset preprocessing, machine learning time series evaluation, time series missing values, time-series-ml, benchmarking, missing-data

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