---
id: datasetsforecast
version: "1.0.1"
license: MIT License
license_treatment: permissive
maintenance: active
---

# datasetsforecast — Datasets for Time series forecasting

License: permissive · Maintenance: active · Downloads: 99.7K/mo

## What it is and what it does

datasetsforecast is a data-loading library that provides programmatic access to standard time-series forecasting benchmarks. It wraps several well-known datasets (Favorita, M3, M4, M5, Hierarchical, Longhorizon, PHM2008) and handles remote fetching and local caching, so you can load them into pandas DataFrames with a single function call. Each dataset module exposes a `load()` method that accepts a local directory and a group identifier, returning train and test splits ready for model evaluation.

The package is designed for researchers and practitioners building and benchmarking forecasting models. It depends on numpy, pandas, scikit-learn, requests, aiohttp, tqdm, xlrd, and utilsforecast—all widely used libraries. The project is actively maintained, supports Python 3.10 through 3.14, and carries a permissive MIT license. It is marked Pre-Alpha, meaning the API may change, but it has been in use since 2022 and receives regular updates.

Use it for:

- Benchmark a new forecasting model against standard datasets like M4 or M5 without manual download and preprocessing.
- Build hierarchical forecasting pipelines using the Hierarchical dataset to test reconciliation methods.
- Evaluate retail demand forecasting on Favorita data for supply-chain optimization projects.
- Compare long-horizon forecasting approaches using the Longhorizon dataset.
- Prototype prognostics and health management models with PHM2008 industrial sensor data.

## Worth the install?

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

Downloads and loads curated time-series forecasting datasets (Favorita, M3, M4, M5, Hierarchical, Longhorizon, PHM2008) into pandas DataFrames, fetching from remote storage on first use.

Yes, if you are building or benchmarking time-series forecasting models. The package eliminates manual dataset acquisition and provides standardized splits for reproducible research. Low install friction, active maintenance, and no known vulnerabilities make it a safe choice. Pre-Alpha status is not a blocker for research use, but be aware the API may change in future releases.

## Install

pip install datasetsforecast
uv add datasetsforecast
poetry add datasetsforecast

## Installing datasetsforecast

Before you install: Low friction: pure Python wheel, 8 common dependencies (numpy, pandas, scikit-learn, requests, aiohttp, tqdm, xlrd, utilsforecast). Active maintenance with recent commits; marked Pre-Alpha but in use.

License in practice: MIT License (permissive): you may use, modify, and distribute this package freely in commercial and private projects with minimal restrictions.

Quickstart:

pip install datasetsforecast

from datasetsforecast.phm2008 import PHM2008
train_df, test_df = PHM2008.load(directory='data', group='FD001')

Requires Python >=3.10. First load will download dataset to specified directory; ensure sufficient disk space and network access.

Verify before relying:

- Total size of all datasets and typical download time not specified in fact sheet.
- Whether datasets are cached after first download or re-fetched on each load.
- API stability and backward compatibility guarantees given Pre-Alpha status.

## Package facts

- License: MIT License (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 99.7K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

time series forecasting datasets, download forecasting benchmarks, M3 M4 M5 dataset loader, hierarchical time series data, favorita retail forecasting, phm2008 prognostics dataset, load forecasting data pandas, time-series, benchmark-datasets, forecasting

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