Parameter Efficient Deep Probabilistic Forecasting (PEDPF) is a repository containing code to run experiments for several deep learning based probabilistic forecasting methods. For more details, see our paper.

First, you need to download the necessary data files.

• UCI Electricity Download and extract LD2011_2014.txt to the folder data\uci_electricity\ (create if necessary).
• UCI Traffic Download and extract to the folder data\uci_traffic\. Run create_df.py to create the dataset from the source files.
• Kaggle Favorita Download and extract all files to the folder data\kaggle_favorita and run prepare_favorita_v3.py (NB: there is some code missing here, this needs to be fixed).
• Kaggle Webtraffic Download and extract all files to the folder data\kaggle_webtraffic and run prepare_webtraffic.py.

Then, run experiments\train.py for the paper's main results. This will sequentially run all the experiments as listed in the experiments\{dataset_name}\experiments_{dataset_name}.csv file. Hence, to change parameters or create more experiments, it is easiest to adjust this .csv file.

The other experiments can be run using the variants of experiments\train_{}.py. Note that some of these variants require installing additional dependencies as well as creating new folders manually.

Olivier Sprangers, Sebastian Schelter, Maarten de Rijke. Parameter Efficient Deep Probabilistic Forecasting. Accepted as journal paper to International Journal of Forecasting.

This project is licensed under the terms of the Apache 2.0 license.

This project was developed by Airlab Amsterdam.