Changelog#
All notable changes to chronocratic-datasets will be documented in this file.
The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.
Entries are managed with towncrier; see
changelog.d/ for unreleased changes.
v0.1.0a3 (2026-06-11)#
Changed#
Fixed#
Use
copy=TrueinDataFrame.to_numpy()to avoid a non-writable tensor warning from PyTorch. (#34)
v0.1.0a1 (2026-06-10) — First Alpha Release#
The first pre-release of chronocratic-datasets. This alpha introduces the complete set of time series datasets, a clean and type-safe API, and full PyTorch Lightning integration.
Expect breaking changes before the 1.0 release. Feedback is welcome.
Added#
Forecasting datasets: ETT, Weather, Electricity with LightningDataModule integration
Classification datasets: UCR (univariate) and UEA (multivariate) benchmarks
ForecastingLoaderMode enum:
RAW_SERIES,INPUT_TARGET,INPUT_ONLYClassificationLoaderMode enum:
SAMPLE_ONLY,SAMPLE_LABELForecastingMode enum:
UNIVARIATE,MULTIVARIATEData caching: Automatic NPZ caching for downloaded and preprocessed data
Data scaling: Configurable normalization via scikit-learn scalers
DDP compliance: All data modules work with distributed training strategies
Utility functions: Cache management, feature extraction, ARFF parsing, collation
Package structure: Full
__init__.pywith 49 re-exported public symbolsBSD 3-Clause license
Sphinx documentation with autodoc-generated API reference
Notes#
Namespace is
chronocratic.datasets(installed viachronocratic-datasetson PyPI).Requires Python 3.12+.
Uses PyTorch Lightning as the primary training framework integration.