yongbingding / time_series_transformers Goto Github PK

This directory contains a Pytorch/Pytorch Lightning implementation of transformers applied to time series. We focus on Transformer-XL and Compressive Transformers.

Transformer-XL is described in this paper Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context by Zihang Dai*, Zhilin Yang*, Yiming Yang, Jaime Carbonell, Quoc V. Le, Ruslan Salakhutdinov (*: equal contribution) Preprint 2018.

Part of this code is from the authors at https://github.com/kimiyoung/transformer-xl.

See requirements.txt. All installed via pip instead of conda.

See etf_data_prep.py in data/etf.

modules.py contains the description of all the components of the model. model.py actually builds it.

Everything is built using the pytorch-lightning wrapper which simplifies generating batches, logging to various experiment tracking frameworks (e.g. Neptune or WandB).

notebook_TXL.ipynb is a notebook that can be loaded in Google Colab to run on TPUs. Note that the entire model and run parameters are specified in a class called GlobalState defined in utils.py with a number of default values.