This repository provides the minimal code for running inference of GTT, with small-scale models available for conceptual experimentation. The complete version is undergoing an internal administrative review and will be released at a later date.
pip install -r requirements.txt
cd src
python test_zeroshot.py --gpu [GPUs] --batch_size [BS] --mode [mode] --data [DS] --uni [uni]
Specify mode to one of the following: tiny, small, large.
Specify data to one of the following: m1, m2, h1, h2, electricity, weather, traffic, ill.
Specify uni to 0 or 1, 0: multivariate forecast, 1: univariate forecast
cd experiments
python test_finetune.py --gpu [GPUs] --batch_size [BS] --mode [mode] --data [DS] --uni [uni] --epochs [eps]
It is rather straightforward to use GTT models for zero-shot forecast on your own data (even with only CPUs), check the tutorial.