first (part 2: other directories)

2026-04-16 15:11:25 +03:00
parent 7814e0bc6b
commit 2f0b85a0c7
203 changed files with 59989 additions and 0 deletions
@@ -0,0 +1,109 @@
+
+import os
+import sys
+import torch
+import torch.nn.functional as F
+import pandas as pd
+import numpy as np
+
+# Path alignment
+sys.path.append(os.getcwd())
+sys.path.append(os.path.join(os.getcwd(), 'ai-engine'))
+
+from pipeline.tiered_loader import TieredDataLoader
+from pipeline.sequence_builder import SequenceBuilder
+from models.hybrid_v11 import HybridDeepModel
+from features.odds_history import OddsHistoryEngine
+from features.synthetic_xg import SyntheticXGModel
+
+DEVICE = 'cpu'
+MODEL_PATH = 'ai-engine/models/v11_hybrid_model.pth'
+TARGET_ID = 'en78ih6ec7exnpxcku3xc3das'
+
+def audit():
+    print(f"🕵️  Auditing Match: {TARGET_ID}")
+    
+    # 1. Pipeline Data
+    builder = SequenceBuilder()
+    X, y, meta = builder.build_sequences()
+    
+    # Check if target is in dataset
+    idx_list = meta.index[meta['match_id'] == TARGET_ID].tolist()
+    if not idx_list:
+        print("❌ Match not found in generated sequences. Is it too old or too new?")
+        return
+    
+    idx = idx_list[0]
+    row_meta = meta.iloc[idx]
+    
+    # 2. Features
+    loader = TieredDataLoader()
+    odds_df = loader.load_gold_data([TARGET_ID])
+    eng = OddsHistoryEngine()
+    xg_model = SyntheticXGModel()
+    
+    # Team Mapping
+    unique_teams = meta['team_id'].unique()
+    team_map = {tid: i for i, tid in enumerate(unique_teams)}
+    
+    # 3. Predict exactly like Backtest
+    state = torch.load(MODEL_PATH, map_location=DEVICE)
+    emb_key = 'entity_emb.weight' if 'entity_emb.weight' in state else 'team_embedding.weight'
+    saved_vocab_size = state[emb_key].shape[0]
+    
+    model = HybridDeepModel(num_teams=saved_vocab_size)
+    new_state = {k.replace('team_embedding', 'entity_emb'): v for k, v in state.items()}
+    model.load_state_dict(new_state, strict=False)
+    model.eval()
+    
+    # Data components
+    team_idx = team_map.get(row_meta['team_id'], 0)
+    entities = torch.LongTensor([team_idx, 0]).unsqueeze(0)
+    seq = torch.FloatTensor(X[idx]).unsqueeze(0)
+    
+    # Context (Odds + xG)
+    odds_lookup = {}
+    for _, r in odds_df.iterrows():
+        mid = r['match_id']
+        if mid not in odds_lookup: odds_lookup[mid] = {}
+        if r['category'] == 'Maç Sonucu': odds_lookup[mid][r['selection']] = r['odd_value']
+        elif r['category'] == '2,5 Alt/Üst':
+            if 'Üst' in r['selection']: odds_lookup[mid]['Over'] = r['odd_value']
+            else: odds_lookup[mid]['Under'] = r['odd_value']
+            
+    odds = odds_lookup.get(TARGET_ID, {'1': 1.0, 'X': 1.0, '2': 1.0, 'Over': 1.0, 'Under': 1.0})
+    syn_xg = 1.35 # Placeholder in trainer for xG component if used
+    hist_win_rate = eng.get_feature(row_meta['team_id'], float(odds.get('1', 1.0)))
+    
+    ctx = torch.FloatTensor([
+        float(odds.get('1', 1.0)), float(odds.get('X', 1.0)), float(odds.get('2', 1.0)),
+        float(odds.get('Over', 1.0)), float(odds.get('Under', 1.0)),
+        syn_xg, syn_xg,
+        hist_win_rate
+    ]).unsqueeze(0)
+    
+    with torch.no_grad():
+        logits_res, pred_goals, logits_btts, logits_ht_ft = model(entities, seq, ctx)
+        probs = F.softmax(logits_res, dim=1).numpy()[0]
+        prob_btts = torch.sigmoid(logits_btts).item()
+        probs_ht = F.softmax(logits_ht_ft, dim=1).numpy()[0]
+        
+    print("\n📊 INTERNAL PIPELINE PREDICTION:")
+    print(f"Target Team: {row_meta['team_id']}")
+    print(f"1X2 Probs: Home:{probs[0]:.4f} Draw:{probs[1]:.4f} Away:{probs[2]:.4f}")
+    print(f"BTTS Prob: {prob_btts:.4f}")
+    
+    ht_map = ["1/1", "1/X", "1/2", "X/1", "X/X", "X/2", "2/1", "2/X", "2/2"]
+    top3_ht = np.argsort(probs_ht)[-3:][::-1]
+    print("Top 3 HT/FT:")
+    for idx_ht in top3_ht:
+        print(f"  {ht_map[idx_ht]}: {probs_ht[idx_ht]:.4f}")
+    
+    actual_res = y[idx][0]
+    actual_ht_idx = int(y[idx][3])
+    print(f"\n✅ ACTUAL REALITY:")
+    print(f"Result (Y): {actual_res} (0.0=Away)")
+    print(f"HT/FT Class: {actual_ht_idx} ({ht_map[actual_ht_idx]})")
+
+if __name__ == "__main__":
+    audit()