similarity_analysis moduleΒΆ

similarity_analysis.py. Main script for running experiments on the distribution of the similarities and the prunning processes.

Usage:

python similarity_analysis.py [1] -n [2] -cfg [3] --mean --theo --help

where:

  • [1] : input similarity matrix (unnormalized). The script expects a ‘exp_configuration.ini’ file in the same folder, usually generated when using main.py.
  • [2] -n: number of samples to plot for each class. Defaults to 5.
  • [3] -cfg: provide a custom configuration file to replace ‘exp_configuration.ini’.
  • --mean: if given, plot an average ROC curve for each ground-truth class.
  • --theo: if given, plot the comparison of the distribution against the theoretical model of the corresponding SIC variant.
  • -h, --help

This outputs pdf histograms of the distribution of similarities for several samples across the matrix and for several normalization parameters.