/supervised_learning_jetpdb

Primary LanguagePythonMIT LicenseMIT

This code is associated with the following paper:

@article{Kit_2023,
doi = {10.1088/1361-6587/acb3f7},
url = {https://dx.doi.org/10.1088/1361-6587/acb3f7},
year = {2023},
month = {feb},
publisher = {IOP Publishing},
volume = {65},
number = {4},
pages = {045003},
author = {A Kit and A E Järvinen and L Frassinetti and S Wiesen and JET Contributors},
title = {Supervised learning approaches to modeling pedestal density},
journal = {Plasma Physics and Controlled Fusion},
}

In order to reproduce the table from the paper, you will

  • Data
    • If you have EUROfusion/JET credentials, please contact the authors for location of the jet pedestal database on Heimdall.
    • Save this to data/ which is one level above src/
  • Instal dependencies from requirements.txt
  • run searches for all models and all input spaces (TBD)
    • chmod +x ./search_all_spaces.sh
    • ./search_all_spaces.sh
    • This produces a .csv for each model and input space combination in the results directory.
  • testing models
    • Using the optimal parameters from the above search, set the base configuration for the model of choice in src/configs/{model_name}_base_config.yaml, and run cv_train_test.py.
    • this will print to stdout the MSE, R2, MAP, etc., metrics for the given configuration
    • additionally it will plot and save a resulting figure to /results/base_results/figures