Diffusion based temporal convolutional model for time-series forecasting. Main focus on long term time-series forecasting.
Project structure
- utils
- utils/dl.py - time-series deep learning models in pytorch with some decorators for training / inference
- utils/data.py - functions for datasets processing
- utils/timediffusion.py - TimeDiffusion model class
- results - folder with csv results files of model evaluation on different datasets
- eval_{
MODEL_NAME}.py - python script for full evaluation pipeline on chosen dataset forMODEL_NAME - TimeDIffussion_exchange_rate_tests.ipynb - visualizations of the forecasts
- TD_vis.ipynb - different visualizations