This repository contains code to replicate the results from Chapter 6 of the Dissertation "Quantifying and Interpreting Uncertainty in Time Series Forecasting" by Kaleb Phipps
This repository is structured in a few key folders:
configs: This folder contains the configs for the models trained for each of the data sets.data: This folder contains the data used for the analyses in the chapter.explanations_analysis: This folder contains multiple Jupyter Notebooks which create the explanations shown in Chapter 6.metrics: This folder contains the evaluation metrics implemented in our code.models: This folder contains the torch based neural network models used to generate probabilistic forecasts.pipelines: This folder contains the pipelines which are often used for preprocessing or synthetic data creation.scripts: This folder contains the scripts which need to be run to generate probabilistic forecasts.
Before the proposed approach can be, you need to prepare a Python environment.
Perform the following steps:
- Set up a virtual environment of Python 3.10 using e.g. venv (
python3.10 -m venv venv) or Anaconda (conda create -n env_name python=3.10). - Possibly install pip via
conda install pip. - Install the dependencies with
pip install -r requirements.txt.
We provide the open source data to replicate our results in the folder data. If you want to apply our approach to further data you will need to download this yourself.
If you are interested in running code, you should navigate to the appropriate script in the scripts folder and run
the respective script from there. Running a script will generate trained models and probabilistic forecasts in a Results
folder. To generate the explanations navigate to explanations_analysis and run the appropriate notebook located there.
If you are interested in applying our method to your own data, you will need to create a new script. You can use the
existing scripts in the scripts folder as orientation for any pipeline you create.
This project is supported by the Helmholtz Association’s Initiative and Networking Fund through Helmholtz AI and by the Helmholtz Association under the Program “Energy System Design”.
This code is licensed under the MIT License.