#!/usr/bin/env python3
"""
Benchmark de performance du système d'anonymisation sur le dataset de test.

Analyse les résultats du batch pour générer un rapport de performance.
"""
import sys
import json
from pathlib import Path
from datetime import datetime
import statistics

def run_baseline_benchmark():
    """Génère le rapport de benchmark à partir des résultats du batch."""
    
    # Répertoires
    baseline_dir = Path("tests/ground_truth/pdfs/baseline_anonymized")
    results_file = baseline_dir / "batch_results.json"
    
    if not results_file.exists():
        print(f"✗ Fichier de résultats non trouvé: {results_file}")
        print(f"  Exécutez d'abord: python3 tools/batch_anonymize_test_dataset.py")
        return 1
    
    # Charger les résultats du batch
    with open(results_file, 'r', encoding='utf-8') as f:
        batch_data = json.load(f)
    
    successful = [r for r in batch_data['results'] if r.get('success')]
    
    if not successful:
        print("✗ Aucun document traité avec succès")
        return 1
    
    print("="*80)
    print("BENCHMARK DE PERFORMANCE - BASELINE")
    print("="*80)
    print(f"\n📅 Date du batch: {batch_data['date']}")
    print(f"📄 Documents: {len(successful)}/{batch_data['total_documents']}")
    print(f"🔍 PII détectés: {batch_data['total_pii']:,}")
    
    # Extraire les métriques
    times = [r['time_s'] for r in successful]
    pii_counts = [r['pii_count'] for r in successful]
    
    # Calculer les statistiques
    stats = {
        "total_documents": len(successful),
        "total_time_s": sum(times),
        "avg_time_s": statistics.mean(times),
        "median_time_s": statistics.median(times),
        "min_time_s": min(times),
        "max_time_s": max(times),
        "stdev_time_s": statistics.stdev(times) if len(times) > 1 else 0.0,
        "total_pii": sum(pii_counts),
        "avg_pii": statistics.mean(pii_counts),
        "median_pii": statistics.median(pii_counts),
        "min_pii": min(pii_counts),
        "max_pii": max(pii_counts),
        "docs_per_second": len(successful) / sum(times),
        "pii_per_second": sum(pii_counts) / sum(times)
    }
    
    # Afficher les statistiques
    print("\n" + "="*80)
    print("STATISTIQUES DE PERFORMANCE")
    print("="*80)
    
    print(f"\n⏱️  Temps de traitement:")
    print(f"   - Total: {stats['total_time_s']:.2f}s")
    print(f"   - Moyen: {stats['avg_time_s']:.2f}s par document")
    print(f"   - Médiane: {stats['median_time_s']:.2f}s")
    print(f"   - Min: {stats['min_time_s']:.2f}s")
    print(f"   - Max: {stats['max_time_s']:.2f}s")
    print(f"   - Écart-type: {stats['stdev_time_s']:.2f}s")
    
    print(f"\n🔍 PII détectés:")
    print(f"   - Total: {stats['total_pii']:,}")
    print(f"   - Moyen: {stats['avg_pii']:.1f} par document")
    print(f"   - Médiane: {stats['median_pii']:.0f}")
    print(f"   - Min: {stats['min_pii']}")
    print(f"   - Max: {stats['max_pii']:,}")
    
    print(f"\n📊 Débit:")
    print(f"   - Documents/seconde: {stats['docs_per_second']:.2f}")
    print(f"   - PII/seconde: {stats['pii_per_second']:.1f}")
    
    # Identifier les documents lents (> 2× moyenne)
    slow_threshold = stats['avg_time_s'] * 2
    slow_docs = [r for r in successful if r['time_s'] > slow_threshold]
    if slow_docs:
        print(f"\n⚠️  Documents lents (> {slow_threshold:.2f}s):")
        for doc in sorted(slow_docs, key=lambda x: x['time_s'], reverse=True)[:5]:
            print(f"   - {doc['pdf']}: {doc['time_s']:.2f}s ({doc['pii_count']} PII)")
    
    # Identifier les documents rapides (< 0.5× moyenne)
    fast_threshold = stats['avg_time_s'] * 0.5
    fast_docs = [r for r in successful if r['time_s'] < fast_threshold]
    if fast_docs:
        print(f"\n⚡ Documents rapides (< {fast_threshold:.2f}s):")
        for doc in sorted(fast_docs, key=lambda x: x['time_s'])[:5]:
            print(f"   - {doc['pdf']}: {doc['time_s']:.2f}s ({doc['pii_count']} PII)")
    
    # Analyser la corrélation PII / temps
    print(f"\n📈 Analyse de corrélation:")
    # Documents avec beaucoup de PII
    high_pii_docs = [r for r in successful if r['pii_count'] > stats['avg_pii'] * 2]
    if high_pii_docs:
        avg_time_high_pii = statistics.mean([r['time_s'] for r in high_pii_docs])
        print(f"   - Documents avec beaucoup de PII (>{stats['avg_pii']*2:.0f}): {len(high_pii_docs)}")
        print(f"     Temps moyen: {avg_time_high_pii:.2f}s")
    
    # Documents avec peu de PII
    low_pii_docs = [r for r in successful if r['pii_count'] < stats['avg_pii'] * 0.5]
    if low_pii_docs:
        avg_time_low_pii = statistics.mean([r['time_s'] for r in low_pii_docs])
        print(f"   - Documents avec peu de PII (<{stats['avg_pii']*0.5:.0f}): {len(low_pii_docs)}")
        print(f"     Temps moyen: {avg_time_low_pii:.2f}s")
    
    # Sauvegarder les résultats
    output_dir = Path("tests/ground_truth/benchmarks")
    output_dir.mkdir(exist_ok=True)
    
    benchmark_data = {
        "date": datetime.now().isoformat(),
        "batch_date": batch_data['date'],
        "configuration": {
            "use_ner": batch_data.get('use_ner', True),
            "use_vlm": batch_data.get('use_vlm', False)
        },
        "statistics": stats,
        "documents": [
            {
                "pdf": r['pdf'],
                "time_s": r['time_s'],
                "pii_count": r['pii_count']
            }
            for r in successful
        ]
    }
    
    json_file = output_dir / "baseline_benchmark.json"
    with open(json_file, 'w', encoding='utf-8') as f:
        json.dump(benchmark_data, f, indent=2, ensure_ascii=False)
    print(f"\n📊 Résultats JSON: {json_file}")
    
    # Export CSV
    csv_file = output_dir / "baseline_benchmark.csv"
    with open(csv_file, 'w', encoding='utf-8') as f:
        f.write("pdf,time_s,pii_count\n")
        for r in successful:
            f.write(f"{r['pdf']},{r['time_s']},{r['pii_count']}\n")
    print(f"📊 Résultats CSV: {csv_file}")
    
    # Vérifier les objectifs de performance
    print("\n" + "="*80)
    print("VALIDATION DES OBJECTIFS")
    print("="*80)
    
    target_time_no_vlm = 10.0  # < 10s par PDF (sans VLM)
    target_time_with_vlm = 30.0  # < 30s par PDF (avec VLM)
    
    # On n'a pas utilisé le VLM dans le batch
    target = target_time_no_vlm
    use_vlm = batch_data.get('use_vlm', False)
    
    if use_vlm:
        target = target_time_with_vlm
    
    print(f"\n🎯 Objectif: < {target}s par document (VLM: {'✓' if use_vlm else '✗'})")
    
    if stats['avg_time_s'] <= target:
        print(f"✅ Temps moyen atteint: {stats['avg_time_s']:.2f}s ≤ {target}s")
    else:
        print(f"⚠️  Temps moyen non atteint: {stats['avg_time_s']:.2f}s > {target}s")
        print(f"   Écart: +{stats['avg_time_s'] - target:.2f}s ({(stats['avg_time_s']/target - 1)*100:.1f}%)")
    
    if stats['max_time_s'] <= target * 3:
        print(f"✅ Temps max acceptable: {stats['max_time_s']:.2f}s ≤ {target * 3}s")
    else:
        print(f"⚠️  Temps max trop élevé: {stats['max_time_s']:.2f}s > {target * 3}s")
    
    # Pourcentage de documents dans l'objectif
    docs_in_target = sum(1 for r in successful if r['time_s'] <= target)
    pct_in_target = (docs_in_target / len(successful)) * 100
    print(f"\n📊 Documents dans l'objectif: {docs_in_target}/{len(successful)} ({pct_in_target:.1f}%)")
    
    if pct_in_target >= 80:
        print(f"✅ Objectif de couverture atteint (≥80%)")
    else:
        print(f"⚠️  Objectif de couverture non atteint (<80%)")
    
    print("\n" + "="*80)
    
    return 0


if __name__ == "__main__":
    sys.exit(run_baseline_benchmark())