"""Porownanie punktacji feedu Edge: grupa nowa vs kontrolna (stara punktacja).

Przydzial do grupy jest staly po hashu id urzadzenia (`app.clips_feed.uses_v1_scoring`),
wiec wystarczy pogrupowac zdarzenia z okresu po wdrozeniu. Liczymy tylko slot 'affinity',
bo tylko tam punktacja dziala (explore losuje tak samo w obu grupach). Srednie sa po
urzadzeniach (kazde urzadzenie z >= MIN_VIEWS obejrzeniami waży tyle samo), zeby kilku
najaktywniejszych nie przesadzalo o wyniku.

  docker compose exec -T worker python3 /srv/scripts/edge_ab_report.py [od] [do]
Domyslnie od wdrozenia (2026-10-02 10:00 UTC) do teraz. Okno sprzed wdrozenia to test A/A:
obie grupy mialy wtedy te sama punktacje, wiec roznice pokazuja sam szum.
"""
from __future__ import annotations

import statistics
import sys
from collections import defaultdict

from sqlalchemy import text

from app.clips_feed import uses_v1_scoring
from app.db import session_scope

SINCE = sys.argv[1] if len(sys.argv) > 1 else "2026-10-02T10:00:00+00:00"
UNTIL = sys.argv[2] if len(sys.argv) > 2 else "2100-01-01T00:00:00+00:00"
MIN_VIEWS = 20
VIEW_KINDS = ("complete", "skip", "impression", "autoswipe")

with session_scope() as s:
    rows = s.execute(text("""
        SELECT device_id, kind, dwell_ms FROM clip_feed_events
        WHERE created_at >= CAST(:since AS timestamptz) AND created_at < CAST(:until AS timestamptz)
          AND slot = 'affinity'
    """), {"since": SINCE, "until": UNTIL}).all()

per = defaultdict(lambda: {"views": 0, "complete": 0, "skip": 0, "like": 0, "not_interested": 0, "dwell": []})
for dev, kind, dwell in rows:
    d = per[dev]
    if kind in VIEW_KINDS:
        d["views"] += 1
        if dwell:
            d["dwell"].append(dwell)
    if kind in d and kind != "views":
        d[kind] += 1

groups: dict[str, list[dict]] = {"nowa": [], "kontrolna": []}
for dev, d in per.items():
    if d["views"] >= MIN_VIEWS:
        groups["kontrolna" if uses_v1_scoring(dev) else "nowa"].append(d)

print(f"od {SINCE} do {UNTIL}, zdarzen {len(rows)}, urzadzen z >= {MIN_VIEWS} obejrzeniami: "
      + ", ".join(f"{g} {len(v)}" for g, v in groups.items()))
for g, devs in groups.items():
    if not devs:
        continue
    def avg(f):
        return 100 * statistics.mean(f(d) for d in devs)
    print(f"  {g:9s} dokonczone {avg(lambda d: d['complete'] / d['views']):5.1f}%"
          f"  lajki {avg(lambda d: d['like'] / d['views']):4.1f}%"
          f"  ucieczki {avg(lambda d: d['skip'] / d['views']):5.1f}%"
          f"  nie-interesuje {avg(lambda d: d['not_interested'] / d['views']):4.1f}%"
          f"  mediana dwell {statistics.median(statistics.median(d['dwell']) for d in devs if d['dwell']) / 1000:.1f} s")