This repository is an implementation of the paper Simultaneously Reconciled Quantile Forecasting of Hierarchically Related Time Series, AISTATS 2021.

Under SHARQ_main, run SHARQ using the following command:

python run_sharq.py

To change the reconciliation method, assign TRAINING_METHOD with different input string. The list of available method is

[‘sharq’, ‘base’, ‘mint_shr’, ‘mint_sam’, ‘mint_ols’, ‘erm’, ‘BU’].

To change the forecasting algorithm and its hyper-parameters, assign MODEL_HYPER_PARAMS with a dictionary that contains model information, for example:

MODEL_HYPER_PARAMS = {‘alg’: ‘rnn’, ‘num_epoch’: 1000, ‘lr’: 0.1, ‘hidden_dim’: 5, ‘layer_dim’: 2,
‘nonlinearity’: ‘tanh’}.

You can either specify all the above hyper-parameters or only one of them.

If you have further questions on implementation details, please contact aaronhan223@utexas.edu