HiReD: Hierarchically Regularized Deep Forecasting
--------------------------------------------------

ICLR 22 supplementary material - Experiment code
------------------------------------------------

The code was tested with the following versions:
Python 3.8.6
Tensorflow 2.2
scikit-learn 0.23.2
pandas 1.1.4
tqdm 4.51.0

All experiments were run on a Titan Xp GPU with 12GB of memory.


Steps to run the experiments:
1. Download the datasets and extract them inside the respective
sub-directories inside the data directory.
Links:
    a) Favorita: https://www.kaggle.com/c/favorita-grocery-sales-forecasting
    b) M5: https://www.kaggle.com/c/m5-forecasting-accuracy
    c) Tourism: https://robjhyndman.com/data/TourismData_v3.csv

2. Set pythonpath:
    source set_pythonpath.sh

3. Preprocess datasets:
    python preprocess/favorita.py
    python preprocess/m5.py
    python preprocess/tourism.py

4. Run the experiments using scripts in the run_scripts directory.
An example command to run the file favorita_reg.sh:
    bash run_scrips/favorita.sh

NOTE: Set the GPU id that you want to use at the beginning of each
run script.

---------------------------------------------------
The descriptions of the run_scripts are as follows:

    favorita.sh: HiReD on Favorita dataset.
    m5.sh: HiReD on M5 dataset.
    tourism.sh: HiReD on Tourism dataset.

---------------------------------------------------
Results: the results will be printed on the terminal for each run.

