Source used to create XYONIX blog for time series forecasting of seasonal influenza via Neural Basis Expansion Analysis for interpretable Time Series forecasting (N-BEATS).
- DATA
- Influenza like illness data: https://www.cdc.gov/flu/weekly/index.htm
- XYONIX
- N-BEATS
- article: https://arxiv.org/abs/1905.10437
- source: https://github.com/philipperemy/n-beats
The blog was created using Python 3.8.0, which can be obtained from https://www.python.org/downloads/. Once you have Python3 installed, issue the following shell commands:
cd seasonal-influenza-xyonix-blog
make clean install run-jupyter
These commands will download and install the N-BEATS source into a build/src subdirectory and install the required Python modules into a build/venv/xyonx-flu virtual environment. After the notebook it launched, select the xyonix-flu kernel as illustrated below:
After installation, you will only need to issue make run-jupyter to continue experimenting with the Jupyter notebook. You can uninstall N-BEATS and delete the virtual environment by running make clean.
Run the cells in the flu-foreacasting.ipynb Jupyter notebook to train and evaluate selected N-BEATS state models. Near the end of the notebook, you should be able to create error heatmaps, like the one shown below, which illustrate the
efficacy of the model over a range of forecast horizons.
