rerun_compare.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""Re-run trifecta predictions with alternative scoring params and compare
against the pasted actual results for 2026-08-10.

Model A  = the predictions already published in predictions-2026-08-10.json
           (open-source TrifectaPredictor, default weights).
Variant B = form-heavy reweight (recent form boosted, barrier/prize damped).
Variant C = place-heavy reweight (place% + prize emphasised, barrier half).

All variants score the SAME runner data (from the prediction file's `form`
block) so the only variable is the scoring function. This is honest
backtesting, not cherry-picking.
"""
from __future__ import annotations

import json
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent))

from src.prediction_model import Runner, Race, TrifectaPredictor

DATA = Path(__file__).resolve().parent / "data"
PRED_FILE = DATA / "predictions" / "predictions-2026-08-10.json"

# --- Actuals from the user paste (1st,2nd,3rd finishing order) ---------------
# Cells with a colon are race times -> no finishing order captured.
ACTUAL = {
    ("Dubbo", "1"): [1, 7, 9], ("Dubbo", "2"): [8, 3, 6], ("Dubbo", "3"): [5, 1, 4],
    ("Dubbo", "4"): [4, 9, 6], ("Dubbo", "5"): [2, 8], ("Dubbo", "6"): [13, 11, 9],
    ("Dubbo", "7"): None,
    ("Kilcoy", "1"): [7, 8], ("Kilcoy", "2"): [6, 9, 4], ("Kilcoy", "3"): [2, 4],
    ("Kilcoy", "4"): [6, 1, 4], ("Kilcoy", "5"): [5, 11, 9], ("Kilcoy", "6"): [10, 4],
    ("Kilcoy", "7"): [15, 9, 1],
    ("Nowra", "1"): [6, 8, 1], ("Nowra", "2"): [9, 11, 1], ("Nowra", "3"): [3, 5, 1],
    ("Nowra", "4"): [4, 11, 2], ("Nowra", "5"): [8, 11, 3], ("Nowra", "6"): [7, 1, 6],
    ("Nowra", "7"): [2, 3, 9],
}

# --- Alternative scoring variants -------------------------------------------
class VariantPredictor(TrifectaPredictor):
    """TrifectaPredictor with configurable component weights."""

    def __init__(self, w=None):
        self.w = {
            "form": 1.0, "win_pct": 1.0, "place_pct": 1.0, "track": 1.0,
            "distance": 1.0, "condition": 1.0, "barrier": 1.0, "prize": 1.0,
        }
        if w:
            self.w.update(w)

    def _score_runner(self, runner):
        w = self.w
        score = 0.0
        form = str(runner.form or runner.last20Starts or "")
        recent = form[-5:] if len(form) > 5 else form
        score += min((recent.count("1") * 8 + recent.count("2") * 4 + recent.count("3") * 4), 25) * w["form"]

        overall = runner.stats.get("overall", {})
        starts = overall.get("starts", 0) or 0
        win_pct = overall.get("winPercent", 0) or 0
        place_pct = overall.get("placePercent", 0) or 0
        score += win_pct * 20 * w["win_pct"]
        score += place_pct * 10 * w["place_pct"]

        track_stats = runner.stats.get("track", {})
        track_starts = track_stats.get("starts", 0) or 0
        track_places = track_stats.get("places", 0) or 0
        score += min((track_places / max(track_starts, 1)) * 10, 10) * w["track"]

        dist_stats = runner.stats.get("distance", {})
        dist_starts = dist_stats.get("starts", 0) or 0
        dist_places = dist_stats.get("places", 0) or 0
        score += min((dist_places / max(dist_starts, 1)) * 8, 8) * w["distance"]

        cond_stats = runner.stats.get("conditions", {})
        for data in cond_stats.values():
            c_starts = data.get("starts", 0) or 0
            c_places = data.get("places", 0) or 0
            score += min((c_places / max(c_starts, 1)) * 8, 8) * w["condition"]

        try:
            barrier = int(runner.barrier) if runner.barrier else 5
            score += max(0, 5 - abs(barrier - 5)) * w["barrier"]
        except Exception:
            score += 3 * w["barrier"]

        try:
            prize = float(str(runner.careerPrizeMoney).replace("$", "").replace(",", ""))
            score += min(prize / 20000, 5) * w["prize"]
        except Exception:
            pass

        return min(round(score, 1), 100)


VARIANTS = {
    "Variant B (form-heavy)": VariantPredictor({
        "form": 1.6, "win_pct": 1.0, "place_pct": 0.8,
        "track": 1.0, "distance": 1.2, "condition": 1.0,
        "barrier": 0.3, "prize": 0.5,
    }),
    "Variant C (place-heavy)": VariantPredictor({
        "form": 0.7, "win_pct": 1.0, "place_pct": 1.7,
        "track": 1.2, "distance": 1.0, "condition": 1.2,
        "barrier": 0.5, "prize": 1.3,
    }),
}


def parse(s):
    return [int(x) for x in s.split("-")]


def build_race(form) -> Race:
    runners = []
    for r in form.get("runners", []):
        runners.append(Runner(
            number=r.get("number"), name=r.get("name", ""),
            jockey=r.get("jockey", ""), trainer=r.get("trainer", ""),
            weight=r.get("weight"), barrier=r.get("barrier"),
            form=r.get("form", ""), last20Starts=r.get("last20Starts", ""),
            careerPrizeMoney=r.get("careerPrizeMoney", "$0"),
            stats=r.get("stats", {}),
        ))
    return Race(
        date=form.get("date"), track=form.get("track"), track_slug=form.get("track_slug", ""),
        race_number=str(form.get("raceNumber")), race_name=form.get("raceName", ""),
        distance=form.get("distance", ""), condition=form.get("condition", ""),
        weather=form.get("weather", ""), race_class=form.get("raceClass", ""),
        start_time=form.get("startTime", ""), prize_money=form.get("prizeMoney", ""),
        number_of_runners=form.get("numberOfRunners", 0), runners=runners,
    )


def score_predictions(pred_map) -> dict:
    """pred_map: (track,rn) -> {'primary','secondary','value','top3'(list of nums)}"""
    tot_exact = tot_box = tot_win = tot_q = races = 0
    full3 = two3 = win_in_top3 = races3 = 0
    details = []
    for (track, rn), a in ACTUAL.items():
        p = pred_map.get((track, rn))
        if not p:
            continue
        prim = parse(p["primary"]); sec = parse(p["secondary"] or ""); val = parse(p["value"] or "")
        top3 = p["top3"]
        if a is None:
            details.append((track, rn, None, p, "time-only"))
            continue
        races += 1
        exact = any(t == a[:3] for t in [prim, sec, val])
        box = any(set(t[:3]) == set(a[:3]) for t in [prim, sec, val] if len(t) >= 3)
        win = a[0] in {prim[0], sec[0], val[0]}
        q = (len(a) == 2 and any(sorted(t[:2]) == sorted(a) for t in [prim, sec, val]))
        tot_exact += exact; tot_box += box; tot_win += win; tot_q += q
        if len(a) == 3:
            races3 += 1
            hits = len(set(a) & set(top3))
            if hits == 3:
                full3 += 1
            elif hits == 2:
                two3 += 1
        if a[0] in top3:
            win_in_top3 += 1
        tag = []
        if exact: tag.append("EXACT")
        if box: tag.append("BOX")
        if win: tag.append("WIN")
        if q: tag.append("QUIN")
        details.append((track, rn, a, p, ",".join(tag) if tag else "miss"))
    summary = {
        "races": races, "exact": tot_exact, "box": tot_box,
        "winner": tot_win, "quinella": tot_q,
        "races3": races3, "full3_in_top3": full3, "two3_in_top3": two3,
        "win_in_top3": win_in_top3, "n_all": sum(1 for v in ACTUAL.values() if v),
        "details": details,
    }
    return summary


def main():
    data = json.loads(PRED_FILE.read_text())
    # Model A: read stored predictions verbatim
    pred_a = {}
    for r in data["races"]:
        p = r["prediction"]
        pred_a[(r["track"], str(r["race_number"]))] = {
            "primary": p["primary"], "secondary": p.get("secondary"),
            "value": p.get("value"), "top3": [t["number"] for t in p["top3"]],
        }
    summary_a = score_predictions(pred_a)

    # Build races once
    races = [build_race(r["form"]) for r in data["races"]]
    results = {"Model A (default)": summary_a}
    out_variants = {}
    for name, predictor in VARIANTS.items():
        pred_map = {}
        for rc in races:
            key = (rc.track, rc.race_number)
            if key not in ACTUAL or ACTUAL[key] is None:
                continue
            pr = predictor.predict(rc)
            if "error" in pr:
                continue
            pred_map[key] = {
                "primary": pr["primary"], "secondary": pr.get("secondary"),
                "value": pr.get("value"), "top3": [t["number"] for t in pr["top3"]],
            }
        summary = score_predictions(pred_map)
        results[name] = summary
        # Save variant predictions for transparency
        out_variants[name] = {f"{k[0]}_R{k[1]}": v for k, v in pred_map.items()}

    # Persist variant predictions
    for name, mp in out_variants.items():
        slug = name.split("(")[0].strip().replace(" ", "_").lower()
        out = DATA / "predictions" / f"predictions-2026-08-10-{slug}.json"
        out.write_text(json.dumps(mp, indent=2))
        print(f"wrote {out}")

    # Print scorecards
    print("\n" + "=" * 78)
    print("SCORECARD COMPARISON  (2026-08-10, 20 scorable races)")
    print("=" * 78)
    for name, s in results.items():
        print(f"\n{name}")
        print(f"  exact trifecta ... {s['exact']}/{s['races']}  ({100*s['exact']/s['races']:.0f}%)")
        print(f"  any box-of-3 .... {s['box']}/{s['races']}  ({100*s['box']/s['races']:.0f}%)")
        print(f"  winner picked ... {s['winner']}/{s['races']}  ({100*s['winner']/s['races']:.0f}%)")
        print(f"  quinella (2-no) . {s['quinella']}")
        print(f"  winner in top3 .. {s['win_in_top3']}/{s['n_all']}  ({100*s['win_in_top3']/s['n_all']:.0f}%)")
        if s['races3']:
            print(f"  full 3 in top3 ... {s['full3_in_top3']}/{s['races3']}   | 2-of-3 in top3 {s['two3_in_top3']}/{s['races3']}")

    # Race-by-race comparison table
    print("\n" + "=" * 78)
    print("RACE-BY-RACE  (actual | A | B | C)")
    print("=" * 78)
    keys = sorted(ACTUAL.keys(), key=lambda k: (k[0], int(k[1])))
    smap = {n: results[n] for n in results}
    for key in keys:
        track, rn = key
        a = ACTUAL[key]
        astr = "-".join(map(str, a)) if a else "(time)"
        row = []
        for n in ["Model A (default)", "Variant B (form-heavy)", "Variant C (place-heavy)"]:
            det = next((d for d in smap[n]["details"] if d[0] == track and d[1] == rn), None)
            if det and det[3]:
                row.append(det[3]["primary"])
            else:
                row.append("--")
        verdict = ""
        # mark which models hit winner
        for n in ["Model A (default)", "Variant B (form-heavy)", "Variant C (place-heavy)"]:
            det = next((d for d in smap[n]["details"] if d[0] == track and d[1] == rn), None)
            if det and det[4] not in (None, "time-only", "miss"):
                verdict += f" {n.split(' ')[1]}:{det[4]}"
        print(f"  {track:7s} R{rn}  {astr:8s} | {row[0]:7s} {row[1]:7s} {row[2]:7s}{verdict}")

    # Save machine-readable results for the PDF builder
    (DATA / "predictions" / "_rerun_summary.json").write_text(json.dumps({
        "date": "2026-08-10",
        "results": {n: {k: v for k, v in s.items() if k != "details"} for n, s in results.items()},
        "details": {n: s["details"] for n, s in results.items()},
    }, indent=2, default=str))
    print("\nwrote data/predictions/_rerun_summary.json")


if __name__ == "__main__":
    main()