https://github.com/norskregnesentral/papershapleyvaluesimprovingkernelshap
Relevant code to the "Improving the Weighting Strategy in KernelSHAP" by Lars Henry Berge Olsen and Martin Jullum
https://github.com/norskregnesentral/papershapleyvaluesimprovingkernelshap
Last synced: 5 months ago
JSON representation
Relevant code to the "Improving the Weighting Strategy in KernelSHAP" by Lars Henry Berge Olsen and Martin Jullum
- Host: GitHub
- URL: https://github.com/norskregnesentral/papershapleyvaluesimprovingkernelshap
- Owner: NorskRegnesentral
- Created: 2025-03-31T20:04:49.000Z (6 months ago)
- Default Branch: main
- Last Pushed: 2025-04-01T19:03:30.000Z (6 months ago)
- Last Synced: 2025-05-05T09:28:10.087Z (5 months ago)
- Language: R
- Size: 33.7 MB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# PaperShapleyValuesImprovingKernelSHAP
Relevant code to the "Improving the Weighting Strategy in KernelSHAP" paper by Lars Henry Berge Olsen and Martin Jullum.
There is a three (or four) step procedure to replicate Figure 2 in the paper.
1. (Optional) Run the file `Run_model.R` to create the model and compute the true Shapley values. This is very time consuming and the output is saved in the `rds` folder.
2. Run the file `Run_sampling.R` with the desired number of repetitions `B` to generate files that generates and store the sampled coalitions using the different sampling strategies.
3. Run the file `Run_SV.R` to compute the MAE error between the estimated and true Shapley values for different number of used unique coalitions, and store the results as data tables to disk.
4. Run the file `Run_plot.R` to load the data tables with the MAE and make the plot.
Note. We have only made the models and true Shapley values available for the `M=10` simulation study, as the files are several GBs for the `M = 20` setting and it thus not suitable for GitHub.