Extra simulations for the paper "Interpretable Deep Image Classification using Rationally Inattentive Utility Maximization" done to validate the findings on newer datasets and architectures.
Instructions on how to install and set up the project. Required dependencies:
torch
numpy
matplotlib
pandas
seaborn
scipy
scikit-learn
huggingface/transformers
tqdm
huggingface/datasets
Directory structure:
- 'cm' -> Confusion matrices for both the experiments
- 'utils' -> Reconstructed utilities for the experiments
- 'plots' -> Plots for the experiments
Files:
- 'generate_confusion_matrix_epoch.py' -> Generates confusion matrices for the experiment with epochs as decision problem
- 'generate_confusion_matrix_variance.py' -> Generates confusion matrices for the experiment with variance as decision problem
- 'optimize_maxmargin.py' -> Reconstructs the max-margin utility for the experiments
- 'optimize_sparse.py' -> Reconstructs the sparse utility for the experiments
- 'generate_plots.py' -> Generates plots for the experiments