reinforcement-learning-experiments/MtMs_sinusoidal_task

MtMs - demonstration on the sinusoidal task

This repository contains the refactored MtMs model used in the prediction part of the M6 forecasting competition. All core functions can be found in the folder src/MtMs. File src/main.R contains a demonstration of how the MtMs model can be used in practice. For more details about the model, please read the accompanying PDF.

To illustrate how it could be used to solve meta-learning and multi-task problems, we apply it to the sinusoidal regression task, which is commonly used to benchmark different meta-learning approaches. To replicate this demonstration:

1. Install dependencies in requirements.txt
2. Run the experiment via either:
   • If you are a DVC user, simply run dvc exp run in the console or run the run.bat file.
   • If not, you can also run directly the src/main.R.

After the experiment is finished, you can find the outputs located in the folder outputs. To modify experiment settings, you can change the contents of the par.yaml file.