NicoWeio / Bachelor-code

Structure

• a_data_selection.py
  • reads the full MC data
  • drops some rows
  • replaces NaNs with an extreme value
  • selects columns according to a_data_selection_features.csv
    • 90% reduction in file size
    • → subsequent analysis is faster
  • outputs to build_large/data.csv
• b_prepare_data.py
  • drops energies outside of a certain range
  • discretizes the energies
  • applies StandardScaler
  • does not write to disk; instead, it's intended to be invoked by other Python files
• c_corn.py
  • provides the CORN classifier as a sklearn classifier
  • ca_corn_functions.py
    • provides helper functions like loss and proba_from_logits
• c_dsea.py
  • …
• d_evaluate.py
  • evaluates the classifier's performance
  • da_evaluate_plots.py
    • outputs plots to build/plots
• x_config.py
  • defines the configuration for the experiment
• x_run.py
  • runs the experiment

TODOs

• MAE → EMD (Wasserstein distance)?
• hyperparameter search?

Requirements

• plotly
  • for passing interactive plots wandb.ai

Quirks (& fixes)

• AttributeError: module 'setuptools._distutils' has no attribute 'version'
  • pip install setuptools==59.5.0

Code references

• class LogisticAT
• https://github.com/mirkobunse/CherenkovDeconvolution.py/blob/master/cherenkovdeconvolution/methods/dsea.py
• https://github.com/janjaek/dsea_mord/blob/master/dsea_best_features.ipynb

Literature

• https://raschka-research-group.github.io/coral-pytorch/tutorials/pytorch_lightning/ordinal-corn_cement
• Easy-to-understand slides explaining CORN/CORAL