chronocratic-datasets

Ready-to-use time series datasets for PyTorch Lightning.

Installation

Install the package via pip:

pip install chronocratic-datasets

Note: The PyPI package name uses a hyphen (chronocratic-datasets), but the import uses the chronocratic.datasets namespace:

from chronocratic.datasets import ...

Quick Start

from pathlib import Path

from chronocratic.datasets import ForecastingMode, WeatherDataModule

weather = WeatherDataModule(
    dataset_file_path=Path("data/weather.csv"),
    mode=ForecastingMode.UNIVARIATE,
)
weather.prepare_data()
weather.setup()
train_loader = weather.train_dataloader()

Datasets

Forecasting

• ETT (Electricity Transformer Temperature): ETTh1, ETTh2, ETTm1, ETTm2 — transformer temperature data at hourly and 15-minute intervals
• Weather: Weather and meteorological features from 2012 to 2017
• Electricity: Hourly electricity load data

Classification

• UCR (Univariate): Archive of univariate time series classification datasets
• UEA (Multivariate): Archive of multivariate time series classification datasets

Features

• PyTorch Lightning DataModules — Drop-in LightningDataModule implementations for seamless integration with Lightning training loops
• Automatic caching with atomic writes — Downloaded and processed data is cached locally with atomic file operations to prevent corruption
• DDP-compliant data loading — Workers share cached data correctly under Distributed Data Parallel training
• Multiple forecasting modes — Switch between UNIVARIATE and MULTIVARIATE forecasting configurations
• Built-in scaling — MinMax and Standard scalers applied automatically per dataset conventions
• Type-safe API — Full type hints and Google-style docstrings for IDE autocomplete and static analysis

Documentation

Comprehensive documentation, including API reference, quickstart guides, and contributing instructions, is available at chronocratic-datasets.readthedocs.io.

License

BSD 3-Clause — see LICENSE for the full text.