Installation is most easily performed in a two-step process, with installation of the dependencies followed by the installation of the package itself. Both can easily be done after the repository is cloned.
Clone the repository:
git clone https://github.com/akuhnregnier/empirical-fire-modellingInstall dependencies into a new conda environment called 'empirical-fire-modelling':
cd empirical-fire-modelling
conda env create -f requirements.yamlActivate this new environment before installing the empirical_fire_modelling package itself (while remaining in the same directory as above):
conda activate empirical-fire-modelling
pip install -e .The repository contains Python code to carry out essential operations across multiple sets of variables ('experiments').
This is stored in the src directory.
Additionally, the analysis directory contains code which makes use of the empirical_fire_modelling package in order to run and visualise the various analyses.
- The package runs various analyses and stores their results
- Model fitting
- Variable importance measures
- ALE plots
- 2D ALE plots
- SHAP values
- SHAP interaction values
- SHAP maps
- Storage of results (e.g. trained models, analyses) using Joblib
- Different input arguments are automatically detected by Joblib and trigger recalculation of results
- As a consequence of the caching mechanisms, cached functions should not call other cached functions or use imported variables, since this would fail to detect changed results from previous calculations
- Result visualisation
- Results are not typically visualised interactively
- Scripts to generate graphs and store them for later viewing are contained in
analysis
- Organisation:
- The
empirical_fire_modellingpackage contains common functionality- Calculation of results or analyses
- Data/result visualisation
- The
analysisfolder contains scripts that use the library modules to execute common functionality across all experiments
- The