DSIPTS: unified library for timeseries modelling

[!CAUTION] PLEASE READ CAREFULLY, check the version of PYTORCH LIGHTNING if installed

[!CAUTION] THE DOCUMENTATION, README and notebook are somehow outdated, some architectures are under review, please be patient and wait for the version 2.0.0 if you want a stable package [!CAUTION] Check frequently the CHANGELOG.md file for the updates!

This library allows to:

• load timeseries in a convenient format
• create tool timeseries with controlled categorical features
• load public timeseries
• train a predictive model using different PyTorch architectures
• define more complex structures using Modifiers (e.g. combining unsupervised learning + deep learning)

Disclamer

The original repository is located here but there is a push mirror in gitlab and you can find it here. Depending on the evolution of the library we will decide if keep both or move definitively to github.

Library

The library can now be found also on pip here and in github here. The readme of the library can be found here.

The pip package (experimental) is available here.

The documentation is here

Suite for training models

Here you can find useful code for training and comparing different architectures using Hydra and Omegaconf (mulitprocess, slurm cluster and optuna sweepers).