deep_analysis.py

"""Final optimization — full grid search and per-date analysis."""
import json
from pathlib import Path

import sys
sys.path.insert(0, '/home/brettanthonysjoberg179/trifecta-bro-hf-space')

from trifecta_bro.data.models import RaceModel, RunnerModel
from trifecta_bro.model.scoring import score_runner
from trifecta_bro.model.pace_analysis import classify_pace

DATA_DIR = Path('/home/brettanthonysjoberg179/trifecta-bro-hf-space/data')
CACHE_DIR = DATA_DIR / 'cache'
RESULTS_DIR = DATA_DIR / 'results'


def load_cached_races(date: str) -> list[dict]:
    races = []
    for f in CACHE_DIR.glob('*.json'):
        try:
            with open(f) as fp:
                data = json.load(fp)
            if data.get('date') == date:
                races.append(data)
        except:
            continue
    return races


def load_results(date: str) -> dict:
    result_lookup = {}
    result_path = RESULTS_DIR / f"{date}-ra.json"
    if not result_path.exists():
        result_path = RESULTS_DIR / f"{date}.json"
    if not result_path.exists():
        return result_lookup
    
    with open(result_path) as f:
        rdata = json.load(f)
    
    if isinstance(rdata, dict):
        if 'tracks' in rdata:
            for track, races in rdata['tracks'].items():
                if isinstance(races, dict):
                    for rn, r in races.items():
                        runners = r.get('runners', [])
                        winners = []
                        for runner in runners:
                            pos = runner.get('position')
                            if pos and pos <= 3:
                                winners.append((pos, runner['number']))
                        winners.sort()
                        actual = [w[1] for w in winners]
                        if actual:
                            result_lookup[(track, int(rn))] = actual
                elif isinstance(races, list):
                    for r in races:
                        if 'trifecta' in r:
                            result_lookup[(track, r.get('race', 0))] = [int(str(x).replace('e','')) for x in r['trifecta']]
        else:
            for track, races in rdata.items():
                if isinstance(races, list):
                    for r in races:
                        if 'trifecta' in r:
                            result_lookup[(track, r['race'])] = [int(str(x).replace('e','')) for x in r['trifecta']]
    
    return result_lookup


def score_race(race_data: dict, weights: dict) -> list[int] | None:
    try:
        runners = []
        for r in race_data.get('runners', []):
            if r.get('scratched'):
                continue
            runner = RunnerModel(
                number=r['number'], name=r.get('name', ''),
                jockey=r.get('jockey'), trainer=r.get('trainer'),
                weight=r.get('weight'), barrier=r.get('barrier'),
                age=r.get('age'), sex=r.get('sex'),
                form=r.get('form', ''), last20_starts=r.get('last20Starts', ''),
                stats=r.get('stats', {}),
            )
            runners.append(runner)
        
        if len(runners) < 3:
            return None
        
        race = RaceModel(
            date=race_data.get('date', ''), track=race_data.get('track', ''),
            track_slug=race_data.get('slug', race_data.get('track', '').lower()),
            race_number=race_data.get('raceNumber', 0),
            race_name=race_data.get('raceName', ''),
            distance=race_data.get('distance'),
            condition=race_data.get('condition'),
            race_class=race_data.get('raceClass'),
            abandoned=race_data.get('abandoned', False),
            start_time=race_data.get('startTime'),
            prize_money=str(race_data.get('prizeMoney', '')),
            number_of_runners=race_data.get('numberOfRunners', len(runners)),
            runners=runners,
        )
        
        pace = classify_pace(race, runners)
        scored = []
        for r in runners:
            sc = score_runner(r, race, pace, weights)
            scored.append((r.number, sc['score']))
        
        scored.sort(key=lambda x: x[1], reverse=True)
        return [s[0] for s in scored[:3]]
    except:
        return None


def evaluate_date(date: str, weights: dict) -> dict:
    result_lookup = load_results(date)
    races = load_cached_races(date)
    stats = {"total": 0, "top1": 0, "exact": 0, "box": 0}
    
    for race_data in races:
        track = race_data.get('track', '')
        race_num = race_data.get('raceNumber', 0)
        actual = result_lookup.get((track, race_num))
        if not actual:
            continue
        
        pred = score_race(race_data, weights)
        if not pred:
            continue
        
        stats["total"] += 1
        if pred == actual:
            stats["exact"] += 1
        if set(pred) == set(actual):
            stats["box"] += 1
        if pred[0] == actual[0]:
            stats["top1"] += 1
    
    if stats["total"] > 0:
        stats["top1_pct"] = stats["top1"] / stats["total"] * 100
    return stats


def evaluate_all(dates: list[str], weights: dict) -> dict:
    agg = {"total": 0, "top1": 0, "exact": 0, "box": 0}
    for date in dates:
        stats = evaluate_date(date, weights)
        agg["total"] += stats["total"]
        agg["top1"] += stats["top1"]
        agg["exact"] += stats["exact"]
        agg["box"] += stats["box"]
    agg["top1_pct"] = agg["top1"] / agg["total"] * 100 if agg["total"] > 0 else 0
    return agg


def main():
    dates = ["2026-08-07", "2026-08-08", "2026-08-09", "2026-08-14"]
    
    # Best weights from first pass
    best = {
        "form": 0.20, "class": 0.13, "distance": 0.08, "track": 0.08,
        "track_distance": 0.08, "condition": 0.07, "jockey": 0.08,
        "fitness": 0.05, "barrier": 0.13, "weight": 0.05, "pace": 0.04,
    }
    
    print("=" * 60)
    print("OPTIMIZED WEIGHTS — PER DATE BREAKDOWN")
    print("=" * 60)
    
    total_races = 0
    total_top1 = 0
    
    for date in dates:
        stats = evaluate_date(date, best)
        if stats['total'] > 0:
            total_races += stats['total']
            total_top1 += stats['top1']
            print(f"\n{date}: {stats['total']} races, {stats['top1']} top1 ({stats['top1']/stats['total']*100:.1f}%)")
            
            # Show individual races
            result_lookup = load_results(date)
            races = load_cached_races(date)
            for race_data in sorted(races, key=lambda x: (x.get('track', ''), x.get('raceNumber', 0))):
                track = race_data.get('track', '')
                race_num = race_data.get('raceNumber', 0)
                actual = result_lookup.get((track, race_num))
                if not actual:
                    continue
                
                pred = score_race(race_data, best)
                if not pred:
                    continue
                
                status = "EXACT!" if pred == actual else ("TOP1!" if pred[0] == actual[0] else "")
                if status:
                    print(f"  {track} R{race_num}: Pred {'→'.join(str(x) for x in pred)} | Actual {'→'.join(str(x) for x in actual)} {status}")
    
    print(f"\n{'='*60}")
    print(f"TOTAL: {total_top1}/{total_races} ({total_top1/total_races*100:.1f}%)")
    print(f"{'='*60}")
    
    # Now try to optimize for Aug 14 specifically to understand what's different
    print("\n" + "=" * 60)
    print("AUG 14 DEEP DIVE")
    print("=" * 60)
    
    # Load Aug 14 results
    result_lookup = load_results("2026-08-14")
    races = load_cached_races("2026-08-14")
    
    # For each race, show what the model missed
    for race_data in sorted(races, key=lambda x: (x.get('track', ''), x.get('raceNumber', 0))):
        track = race_data.get('track', '')
        race_num = race_data.get('raceNumber', 0)
        actual = result_lookup.get((track, race_num))
        if not actual:
            continue
        
        pred = score_race(race_data, best)
        if not pred:
            continue
        
        # Show top 3 picks vs actual
        print(f"\n{track} R{race_num}:")
        print(f"  Actual: {'→'.join(str(x) for x in actual)}")
        print(f"  Predicted: {'→'.join(str(x) for x in pred)}")
        
        # Show ranked runners
        runners = []
        for r in race_data.get('runners', []):
            if r.get('scratched'):
                continue
            runner = RunnerModel(
                number=r['number'], name=r.get('name', ''),
                jockey=r.get('jockey'), trainer=r.get('trainer'),
                weight=r.get('weight'), barrier=r.get('barrier'),
                age=r.get('age'), sex=r.get('sex'),
                form=r.get('form', ''), last20_starts=r.get('last20Starts', ''),
                stats=r.get('stats', {}),
            )
            runners.append(runner)
        
        race = RaceModel(
            date='2026-08-14', track=track,
            track_slug=track.lower(),
            race_number=race_num,
            race_name=race_data.get('raceName', ''),
            distance=race_data.get('distance'),
            condition=race_data.get('condition'),
            race_class=race_data.get('raceClass'),
            number_of_runners=race_data.get('numberOfRunners', len(runners)),
            runners=runners,
        )
        
        pace = classify_pace(race, runners)
        
        scored = []
        for r in runners:
            sc = score_runner(r, race, pace, best)
            scored.append({
                'no': r.number, 'name': r.name, 'score': sc['score'],
                'barrier': r.barrier, 'weight': r.weight, 'form': r.form,
                'win_pct': r.win_percent, 'place_pct': r.place_percent,
            })
        
        scored.sort(key=lambda x: x['score'], reverse=True)
        
        print(f"  Ranked:")
        for i, s in enumerate(scored[:6]):
            marker = " <-- WINNER" if s['no'] == actual[0] else ""
            print(f"    {i+1}. #{s['no']} {s['name'][:20]:<20} score={s['score']:.1f} b={s['barrier']} w={s['weight']} form={s['form']} win={s['win_pct']*100:.0f}%{marker}")


if __name__ == "__main__":
    main()