# -*- coding: utf-8 -*-
"""Recompute the repeat-run figures from houston-repeat-run-names.csv.

Each row lists, in order of first mention, the firms one run named for one question. Three runs
per question and engine: the October answer and two later repeats of the same question. Overlap is
the Jaccard share: firms named by both runs divided by firms named by either.

    python repeat_run_score.py houston-repeat-run-names.csv
"""
import csv
import random
import statistics as st
import sys
from collections import defaultdict

PAIRS = {"october~repeat_1": ("october", "repeat_1"), "repeat_1~repeat_2": ("repeat_1", "repeat_2"),
         "october~repeat_2": ("october", "repeat_2")}


def boot_ci(xs, n=4000, seed=20261104):
    rng = random.Random(seed)
    if not xs:
        return (None, None)
    ms = sorted(st.mean(rng.choice(xs) for _ in xs) for _ in range(n))
    return (round(ms[int(.025 * n)], 3), round(ms[int(.975 * n)], 3))


def main(path):
    runs = defaultdict(dict)
    for r in csv.DictReader(open(path, encoding="utf-8", newline="")):
        names = [x.strip() for x in r["firms_named_in_order"].split(" | ") if x.strip()]
        runs[(r["engine"], r["query_id"])][r["run"]] = names
    for eng in sorted({k[0] for k in runs}):
        qs = [v for k, v in runs.items() if k[0] == eng]
        print(f"\n{eng}: {len(qs)} questions")
        for pn, (a, b) in PAIRS.items():
            js, zero, same_first = [], 0, 0
            for q in qs:
                A, B = set(q[a]), set(q[b])
                if not (A | B):
                    continue
                j = len(A & B) / len(A | B)
                js.append(j)
                zero += j == 0
                same_first += bool(q[a]) and bool(q[b]) and q[a][0] == q[b][0]
            print(f"  {pn}: n={len(js)} mean overlap {st.mean(js):.3f} (95% {boot_ci(js)}), median {st.median(js):.3f}, "
                  f"no firm in common {zero / len(js):.3f}, same firm named first {same_first / len(js):.3f}")


if __name__ == "__main__":
    main(sys.argv[1])