forecastlib

A PyTorch Lightning–based library of time series forecasting models, data loaders and utilities. Adapted and reorganized from thuml/Time-Series-Library: the models are more or less the same, wrapped for Lightning with better logging, plus additions for differencing, database-backed model/forecast storage and a Dash dashboard.

Using uv:

uv add "forecastlib @ git+https://github.com/mspils/forecastlib"

or with pip:

pip install "forecastlib @ git+https://github.com/mspils/forecastlib"

uv add "forecastlib[database,wrappers] @ git+https://github.com/mspils/forecastlib"

import forecastlib
from forecastlib import CustomLightningModule

# CustomLightningModule, EnsembleModule, UncertaintyLightningModule, WeightingModel
# are exported at the top level.

When using the database extra, configuration is read from a .env file — see .env.example.

uv sync --extra dev
uv run poe lint
uv run poe test

MIT — see LICENSE.