Skip to content

Time Series Orchestra (TSorchestra) is a novel ensemble framework designed for zero-shot time series forecasting. It is built upon a curated collection of time series foundation models. The architecture is designed to leverage the specialized capabilities of its constituent models to deliver SOTA performance and generalization across datasets.

License

Notifications You must be signed in to change notification settings

DC-research/TSorchestra

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TSorchestra

[Ongoing Project]

Time Series Orchestra (TSorchestra) is a novel ensemble framework designed for zero-shot time series forecasting. It's a curated collection of time series foundation models (TSFMs) that leverages each TSFM's strengths to create something greater than the sum of its parts, yielding SOTA performance.

Set Up


  1. Create a new conda environment named tso from our .yml file:
conda env create -f environment.yml
  1. Download the GIFT-Eval benchmark from Hugging Face:
mkdir data
huggingface-cli download Salesforce/GiftEval --repo-type=dataset --local-dir data
  1. Set up the environment variable for loading the datasets: bash echo "GIFT_EVAL=data" >> .env

Usage


Run our evaluation script to reproduce our results:

chmod +x ./cli/eval.sh
./cli/eval.sh

About

Time Series Orchestra (TSorchestra) is a novel ensemble framework designed for zero-shot time series forecasting. It is built upon a curated collection of time series foundation models. The architecture is designed to leverage the specialized capabilities of its constituent models to deliver SOTA performance and generalization across datasets.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published