summarize-confusion-report

#! /usr/bin/python3

import os
import sys
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), 'lib')))

import collections
import csv
import fileinput

import numpy as np
from titlecase import titlecase

def inner_main(rd, wr):
    header = next(rd)
    assert header[:4] == ["algorithm", "cal_set", "id", "true_rgn"]
    regions = [titlecase(s.replace('_', ' ')) for s in header[4:]]

    matrices = collections.defaultdict(
        lambda: np.zeros((len(regions), len(regions))))

    for row in rd:
        if row[0] == "algorithm":
            assert row == header
            continue

        v = np.array([float(x) for x in row[4:]])
        if not all(np.isfinite(v)):
            sys.stderr.write("skipping: " + repr(row) + "\n")
            continue
        mtx = matrices[row[0], row[1]]
        mtx[int(row[3]),:] += v

    wr.writerow(("algorithm", "cal_set", "rgn1", "rgn2", "confusion"))
    for (alg, cal_set), mtx in matrices.items():
        # normalize rows
        rs = mtx.sum(axis=1)
        rs[rs == 0] = 1
        mtx /= rs[:,np.newaxis]
        for i, a in enumerate(regions):
            for j, b in enumerate(regions):
                wr.writerow((alg, cal_set, a, b, mtx[i,j]))

def main():
    with fileinput.input() as inf, sys.stdout as ouf:
        rd = csv.reader(inf)
        wr = csv.writer(ouf, dialect='unix', quoting=csv.QUOTE_MINIMAL)
        inner_main(rd, wr)

main()