The current repository contains source code that showcases the GRRNN model for Causal Time-series Forecasting. The starting point for optimizing the hyperparameters, training and testing the model is the script 'run.py'. Sample datasets are included in 'datasets' folder. For each dataset there is a corresponding 'source.txt' file with a link from which the datasets were obtained.

Command-line options: --dataset [ETD|SML2010|AirQuality|energyco|poll] --mode [optimize|train|test] --horizon: [N]

Example:

• To optimize the hyperparameters for forecasting 7 steps ahead for dataset 'SML2010' execute: python run.py --dataset SML2010 --horizon 7 --mode optimize The optimal hyperparameters are output to the file 'optimized_configuration.txt'. Note: 'configuration_space.txt' contains the range and initial value for each hyperparameter. The range and initial value can be altered with the latter falling in the former.

• To train the model with the optimized hyperparameters produced in the previous step execute: python run.py --dataset SML2010 --horizon 7 --mode train The train model is output to the file 'model.pt'

• To test the trained model produced in the previous step execute: python run.py --dataset SML2010 --horizon 7 --mode test