Metadata-Version: 2.1
Name: torch-es
Version: 0.0.1
Summary: Double Seasonal Exponential Smoothing using PyTorch + ES-RNN capabilities on top
Home-page: https://github.com/vshulyak/torch-es
Author: Vladimir Shulyak
Author-email: vladimir@shulyak.net
License: UNKNOWN
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Description-Content-Type: text/markdown
Requires-Dist: torch (>=1.0.0)
Requires-Dist: numpy (>=1.16.0)

# torch-es
Double Seasonal Exponential Smoothing using [`PyTorch`](https://pytorch.org) with batched data and multiple series training support.


# 📋 Roadmap

There are lots of tools built on top of the code in this repository, so the plan is to add them here eventually.

Here's what's published:

- [x] 3d Holt-Winters implementation
- [x] Additive and Multiplicative seasonalities
- [x] Blender module to merge predictions from multiple series.
- [ ] Training loop for normal and bptt training.
- [ ] Uncertainty estimation via sampling.
- [ ] Additional losses
- [ ] RNN training on top of HW.

# 📚 Dependencies

- torch
- numpy


