compare_accuracy.py now saves confusion matrices and can be called on the model performance output csv directly
