Forchestra

TS2Vec

We used the TS2Vec code from: https://github.com/yuezhihan/ts2vec. This repo doesn't have a copyright for it.

How to Run (Examples)

How to train a model without initializing

python main.py --model_save_path ./params/forchestra.pt --output_fname_prefix ./results/prediction

How to train a model with initializing

python main.py --repr_model_path ./params/ts2vec.pt --base_model_path ./params/base.pt --model_save_path ./params/forchestra.pt --output_fname_prefix ./results/prediction

How to get prediction using a trained model

python main.py --model_load_path ./params/forchestra.pt --output_fname_prefix ./results/prediction --skip_train

Techinical Notes

• We used different scaling method for representation and prediction module. Selecting a proper scaling method for each module is important.
• CNN-based model works better for the representation module (classification), while LSTM-based model works better for the prediction module (regression).