grading-on-a-curve/code/learning_curves/plot_T1s.py

#!/usr/bin/env python3

import sys
import subprocess
import math
import numpy as np
import matplotlib.pyplot as plt


def get_commit_xps(git_dir, paths):
    command = "git -C " + git_dir + " log " \
        "--full-history --reverse --no-merges --use-mailmap "\
        "--format='format:%ct	%aN <%aE>' -- " + paths + \
        " | sort -n | cut -f2"
    lines = subprocess.check_output(command, shell=True,
                                    universal_newlines=True).strip()
    assert lines
    author_xps = {}
    xps = []
    for line in lines.splitlines():
        author = line.strip()
        if author not in author_xps:
            author_xps[author] = 1
        xps.append(author_xps[author])
        author_xps[author] += 1
    return xps


def find_T1_at_l(commit_xps, num_vccs, l, precision):
    left = 0.0
    right = 1.0
    s = len(commit_xps) - num_vccs
    print("l: {} s: {}".format(l, s))
    while (right - left) > precision:
        T1_guess = (left + right) / 2
        mean = sum([1.0-T1_guess*c**-l for c in commit_xps])
        assert(s > mean)
        assert(math.log(s/mean) > 0)
        # Pr[S>=s] (i.e., the probability that a dist with T1_guess would have produced more vuln-free commits)
        p = math.exp(s - mean - s * math.log(s/mean))
        print(T1_guess, mean, p)
        if p < 0.05:
            # the probability of fewer good commits is <5%,
            # so this T1 would have produced more vulns with >95% prob.
            # we can lower our guess for T1
            right = T1_guess
        else:
            left = T1_guess
    return T1_guess


def main(argv):
    git_dirs = argv[1].split(':')
    paths = argv[2].split(':')
    plot_path = argv[3]
    commit_xps = [xp for i in range(len(git_dirs)) for xp in
                  get_commit_xps(git_dirs[i], paths[i])]
    num_vccs = int(argv[4])
    l_vals = np.arange(0.01, 0.2, 0.02)
    print(l_vals)
    T1s = [find_T1_at_l(commit_xps, num_vccs, l, 0.0001) for l in l_vals]
    print(T1s)
    plt.plot(l_vals, T1s)
    plt.savefig(plot_path)


if __name__ == '__main__':
    main(sys.argv)