File size: 25,107 Bytes
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0c3974e
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
450368b
 
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0c3974e
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d49cb64
 
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d49cb64
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4276a62
 
d49cb64
 
 
4276a62
d49cb64
 
4276a62
d49cb64
4276a62
 
d49cb64
 
 
 
 
4276a62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
"""
Virus Prediction Model Handler Module
Encapsulates TabularResNet model loading, feature preprocessing, and prediction logic
Uses PyTorch-based neural networks with bundled preprocessing in .pth files
FastAPI-compatible version (no Streamlit dependencies)
"""

from pathlib import Path
import logging

import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import LabelEncoder, StandardScaler

# Configure logging
logger = logging.getLogger(__name__)

# ============================================================================
# VIRUS & SYMPTOM MAPPINGS
# ============================================================================

# Main virus mapping (26 classes)
DEFAULT_VIRUS_MAPPING = {
    0: 'Chikungunya Virus',
    1: 'Dengue Virus',
    2: 'Enterovirus',
    3: 'Hepatitis A Virus',
    4: 'Hepatitis B Virus',
    5: 'Hepatitis C Virus',
    6: 'Hepatitis E Virus',
    7: 'Herpes simplex virus',
    8: 'Influenza A H1N1',
    9: 'Influenza A H3N2',
    10: 'Influenza B Victoria',
    11: 'Japanese Encephalitis',
    12: 'Leptospira',
    13: 'Measles Virus',
    14: 'Mumps Virus',
    15: 'OtherViruses',
    16: 'Parvovirus',
    17: 'Respiratory Adenovirus',
    18: 'Respiratory Syncytial Virus RSV',
    19: 'Respiratory Syncytial Virus-A RSV-A',
    20: 'Respiratory Syncytial Virus-B RSV-B',
    21: 'Rotavirus',
    22: 'Rubella',
    23: 'SARS-Cov-2',
    24: 'Scrub typhus Orientia tsutsugamushi',
    25: 'Varicella zoster virus VZV'
}

# Other Virus sub-classification mapping (13 classes)
DEFAULT_OTHER_VIRUS_MAPPING = {
    0: 'HIV',
    1: 'Haemophilus influenzae',
    2: 'Herpes simplex virus (HSV)',
    3: 'Human papillomavirus (HPV)',
    4: 'Kyasanur Forest Disease',
    5: 'Metapneumovirus',
    6: 'Norovirus',
    7: 'Other Influenza',
    8: 'Rhinovirus',
    9: 'Toxoplasma',
    10: 'Unknown',
    11: 'West Nile virus (WNV)',
    12: 'Zika'
}

VIRUS_MAPPING = dict(DEFAULT_VIRUS_MAPPING)
OTHER_VIRUS_MAPPING = dict(DEFAULT_OTHER_VIRUS_MAPPING)
COMBINED_VIRUS_MAPPING = {}


def _read_virus_mapping_csv(csv_path, expected_count=None):
    """Read virus mapping from CSV file"""
    df = pd.read_csv(csv_path)
    required_cols = {"Original", "Encoded"}
    if not required_cols.issubset(df.columns):
        raise ValueError(
            f"Invalid mapping file: {csv_path}. Expected columns: {required_cols}."
        )

    df = df.dropna(subset=["Original", "Encoded"])
    df["Encoded"] = df["Encoded"].astype(int)
    mapping = dict(zip(df["Encoded"], df["Original"].astype(str)))

    if expected_count is not None and len(mapping) != expected_count:
        logger.warning(
            f"Mapping size mismatch for {csv_path}. Expected {expected_count}, got {len(mapping)}."
        )

    return mapping


def refresh_virus_mappings(major_csv_path=None, other_csv_path=None):
    """
    Reload virus name mappings from CSV files and update in place.

    Args:
        major_csv_path: Path to encoding_major_VIRUS_NAME.csv
        other_csv_path: Path to encoding_other_VIRUS_NAME.csv
    """
    base_dir = Path(__file__).resolve().parent
    major_csv_path = major_csv_path or base_dir / "encoding_major_VIRUS_NAME.csv"
    other_csv_path = other_csv_path or base_dir / "encoding_other_VIRUS_NAME.csv"

    major_mapping = dict(DEFAULT_VIRUS_MAPPING)
    other_mapping = dict(DEFAULT_OTHER_VIRUS_MAPPING)

    try:
        if Path(major_csv_path).exists():
            major_mapping = _read_virus_mapping_csv(major_csv_path, expected_count=26)
        else:
            logger.warning(f"Major mapping file not found: {major_csv_path}")
    except Exception as exc:
        logger.warning(f"Failed to load major mapping from CSV: {exc}")

    try:
        if Path(other_csv_path).exists():
            other_mapping = _read_virus_mapping_csv(other_csv_path, expected_count=13)
        else:
            logger.warning(f"Other mapping file not found: {other_csv_path}")
    except Exception as exc:
        logger.warning(f"Failed to load other mapping from CSV: {exc}")

    VIRUS_MAPPING.clear()
    VIRUS_MAPPING.update(major_mapping)

    OTHER_VIRUS_MAPPING.clear()
    OTHER_VIRUS_MAPPING.update(other_mapping)

    COMBINED_VIRUS_MAPPING.clear()
    COMBINED_VIRUS_MAPPING.update(
        {f"main_{k}": v for k, v in VIRUS_MAPPING.items() if k != 15}
    )
    COMBINED_VIRUS_MAPPING.update(
        {f"other_{k}": f"Other Viruses → {v}" for k, v in OTHER_VIRUS_MAPPING.items()}
    )


# Initialize mappings
refresh_virus_mappings()

# All clinical symptoms (no spaces to match training data)
ALL_SYMPTOMS = [
    'HEADACHE', 'IRRITABILITY', 'ALTEREDSENSORIUM', 'SOMNOLENCE', 'NECKRIGIDITY', 'SEIZURES',
    'DIARRHEA', 'DYSENTERY', 'NAUSEA', 'VOMITING', 'ABDOMINALPAIN',
    'MALAISE', 'MYALGIA', 'ARTHRALGIA', 'CHILLS', 'RIGORS', 'FEVER',
    'BREATHLESSNESS', 'COUGH', 'RHINORRHEA', 'SORETHROAT',
    'BULLAE', 'PAPULARRASH', 'PUSTULARRASH', 'MUSCULARRASH', 'MACULOPAPULARRASH', 'ESCHAR',
    'DARKURINE', 'HEPATOMEGALY', 'JAUNDICE',
    'REDEYE', 'DISCHARGEEYES', 'CRUSHINGEYES', 'SWELLINGEYES', 'RETROORBITALPAIN'
]


# ============================================================================
# DEVICE DETECTION
# ============================================================================

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
logger.info(f"Using device: {DEVICE}")

# ============================================================================
# TABULARRESNET ARCHITECTURE
# ============================================================================

class GEGLU(nn.Module):
    """Gated Linear Unit with GELU activation"""
    def __init__(self, d_model, d_ff):
        super().__init__()
        self.fc1 = nn.Linear(d_model, d_ff * 2)
        self.fc2 = nn.Linear(d_ff, d_model)

    def forward(self, x):
        a, b = self.fc1(x).chunk(2, dim=-1)
        return self.fc2(a * F.gelu(b))


class TransformerBlock(nn.Module):
    """Transformer block with gated residual connections"""
    def __init__(self, d_model=128, n_heads=4, d_ff=256, dropout=0.1):
        super().__init__()
        self.ln1 = nn.LayerNorm(d_model)
        self.attn = nn.MultiheadAttention(
            d_model, n_heads, dropout=dropout, batch_first=True
        )
        self.ln2 = nn.LayerNorm(d_model)
        self.ff = GEGLU(d_model, d_ff)

        self.attn_gate = nn.Parameter(torch.zeros(1))
        self.ff_gate = nn.Parameter(torch.zeros(1))
        self.drop = nn.Dropout(dropout)

    def forward(self, x):
        h = self.ln1(x)
        attn_out, _ = self.attn(h, h, h, need_weights=False)
        x = x + torch.sigmoid(self.attn_gate) * self.drop(attn_out)

        h = self.ln2(x)
        ff_out = self.ff(h)
        x = x + torch.sigmoid(self.ff_gate) * self.drop(ff_out)
        return x


class TabularResNet(nn.Module):
    """Enhanced TabularResNet for virus classification"""
    def __init__(self, num_binary, num_continuous, cat_dims,
                 num_classes, d_token=256, depth=2, dropout=0.1):
        super().__init__()

        self.num_cat = len(cat_dims)
        self.num_continuous = num_continuous

        # Categorical Embeddings
        self.cat_embeds = nn.ModuleList([
            nn.Embedding(card, emb) for card, emb in cat_dims
        ])
        self.cat_proj = nn.ModuleList([
            nn.Linear(emb, d_token) for _, emb in cat_dims
        ])

        # Continuous Features
        self.cont_proj = nn.ModuleList([
            nn.Linear(1, d_token) for _ in range(num_continuous)
        ])
        self.cont_scale = nn.ParameterList([
            nn.Parameter(torch.ones(d_token)) for _ in range(num_continuous)
        ])

        # Binary Features
        self.bin_linear = nn.Linear(num_binary, d_token)
        self.bin_gate = nn.Parameter(torch.zeros(1))

        # Token Management
        self.max_tokens = 1 + self.num_cat + num_continuous + (1 if num_binary > 0 else 0)
        self.cls_token = nn.Parameter(torch.zeros(1, 1, d_token))
        self.pos_embed = nn.Parameter(torch.zeros(1, self.max_tokens, d_token))

        # Transformer Blocks
        self.blocks = nn.ModuleList([
            TransformerBlock(d_token, dropout=dropout) for _ in range(depth)
        ])

        self.norm = nn.LayerNorm(d_token)

        # Projection head
        self.proj_head = nn.Sequential(
            nn.Linear(d_token, d_token * 2),
            nn.ReLU(inplace=True),
            nn.Linear(d_token * 2, d_token),
            nn.ReLU(inplace=True),
            nn.Linear(d_token, 128)
        )

        # Classification head
        self.head = nn.Linear(d_token, num_classes)

    def forward(self, xb, xc, xcat, return_embed=False):
        B = xb.size(0)
        tokens = []

        # Categorical Tokens
        for i in range(self.num_cat):
            cat_emb = self.cat_embeds[i](xcat[:, i])
            tokens.append(self.cat_proj[i](cat_emb).unsqueeze(1))

        # Continuous Tokens
        for i in range(self.num_continuous):
            cont_token = self.cont_proj[i](xc[:, i:i+1])
            cont_token = cont_token * self.cont_scale[i]
            tokens.append(cont_token.unsqueeze(1))

        # Binary Token
        if xb.numel() > 0:
            bin_emb = torch.sigmoid(self.bin_gate) * self.bin_linear(xb)
            tokens.append(bin_emb.unsqueeze(1))

        # Combine Tokens
        x = torch.cat(tokens, dim=1) if tokens else torch.randn(B, 1, 256).to(DEVICE)
        cls = self.cls_token.expand(B, -1, -1)
        x = torch.cat([cls, x], dim=1)

        pos_embed_used = self.pos_embed[:, :x.size(1), :]
        x = x + pos_embed_used

        # Transformer Blocks
        for blk in self.blocks:
            x = blk(x)

        x = self.norm(x)

        # Pooling
        pooled = 0.7 * x[:, 0] + 0.3 * x[:, 1:].mean(dim=1)

        # Return Embeddings or Logits
        if return_embed:
            z = F.normalize(self.proj_head(pooled), dim=1)
            return pooled, z

        return self.head(pooled)


# ============================================================================
# VIRUS PREDICTOR CLASS
# ============================================================================

class VirusPredictor:
    """
    Encapsulates TabularResNet model loading, feature preprocessing, and prediction.
    Loads bundled .pth files with model weights + preprocessing objects.
    """
    
    def __init__(self, model1_path='models/CustomMajor.pth', 
                 model2_path='models/CustomOther.pth'):
        """
        Initialize predictor by loading both pretrained models.
        
        Args:
            model1_path: Path to primary model .pth file (26 major viruses)
            model2_path: Path to secondary model .pth file (13 other virus sub-types)
        """
        self.model1 = None
        self.model2 = None
        self.preprocessing1 = None
        self.preprocessing2 = None
        self.load_models(model1_path, model2_path)
    
    def load_models(self, model1_path, model2_path):
        """Load both TabularResNet models with bundled preprocessing"""
        try:
            allowlisted = [
                SimpleImputer,
                StandardScaler,
                LabelEncoder,
                np.core.multiarray._reconstruct,
            ]
            torch.serialization.add_safe_globals(allowlisted)

            def _safe_torch_load(path):
                with torch.serialization.safe_globals(allowlisted):
                    try:
                        return torch.load(path, map_location=DEVICE, weights_only=True)
                    except Exception:
                        import warnings
                        warnings.filterwarnings('ignore', message='.*weights_only.*')
                        return torch.load(path, map_location=DEVICE, weights_only=False)

            # Load Model 1 (Primary - 26 viruses)
            logger.info(f"Loading Model 1 from {model1_path}")
            checkpoint1 = _safe_torch_load(model1_path)
            config1 = checkpoint1['model_config']
            self.model1 = TabularResNet(**config1).to(DEVICE)
            self.model1.load_state_dict(checkpoint1['model_state_dict'])
            self.model1.eval()
            self.preprocessing1 = checkpoint1['preprocessing']
            self._normalize_imputer_state(self.preprocessing1)
            logger.info("Model 1 loaded successfully")
            
            # Load Model 2 (Secondary - Other Viruses)
            logger.info(f"Loading Model 2 from {model2_path}")
            checkpoint2 = _safe_torch_load(model2_path)
            config2 = checkpoint2['model_config']
            self.model2 = TabularResNet(**config2).to(DEVICE)
            self.model2.load_state_dict(checkpoint2['model_state_dict'])
            self.model2.eval()
            self.preprocessing2 = checkpoint2['preprocessing']
            self._normalize_imputer_state(self.preprocessing2)
            logger.info("Model 2 loaded successfully")
            
            return True
            
        except FileNotFoundError as e:
            logger.error(f"Model file not found: {e}")
            return False
        except Exception as e:
            logger.error(f"Error loading models: {e}")
            return False

    @staticmethod
    def _normalize_imputer_state(preprocessing):
        """Backfill SimpleImputer attributes for cross-version sklearn compatibility"""
        for key in ('imp_cont', 'imp_bin'):
            imputer = preprocessing.get(key)
            if isinstance(imputer, SimpleImputer) and not hasattr(imputer, '_fill_dtype'):
                if hasattr(imputer, '_fit_dtype'):
                    imputer._fill_dtype = imputer._fit_dtype
                elif hasattr(imputer, 'statistics_'):
                    imputer._fill_dtype = np.asarray(imputer.statistics_).dtype
                else:
                    imputer._fill_dtype = np.dtype('float64')
    
    def preprocess_features(self, patient_data, preprocessing):
        """Transform patient data dict → binary, continuous, categorical tensors"""
        try:
            binary_cols = preprocessing['binary_cols']
            cat_cols = preprocessing['cat_cols']
            cont_cols = preprocessing['cont_cols']
            imp_cont = preprocessing['imp_cont']
            scaler = preprocessing['scaler']
            imp_bin = preprocessing['imp_bin']
            le_dict = preprocessing['le_dict']
            
            # Create DataFrame
            df = pd.DataFrame([patient_data])
            
            # ========== FEATURE ENGINEERING ==========
            
            # Age features
            age_median = df['age'].median() if 'age' in df.columns else 30
            df['age'] = df['age'].fillna(age_median).clip(0, 120)
            
            age_group = pd.cut(df['age'], bins=[0, 5, 18, 45, 65, 150], labels=[0, 1, 2, 3, 4]).cat.codes
            df['age_group'] = age_group.replace(-1, 2)
            
            # Symptom handling
            symptom_cols = [col for col in ALL_SYMPTOMS if col in df.columns]
            for col in symptom_cols:
                df[col] = df[col].fillna(0)
            
            df['durationofillness'] = df['durationofillness'].fillna(0)
            
            # Symptom groups
            respiratory_cols = ['COUGH', 'BREATHLESSNESS', 'RHINORRHEA', 'SORETHROAT']
            gi_cols = ['DIARRHEA', 'DYSENTERY', 'NAUSEA', 'VOMITING', 'ABDOMINALPAIN']
            neuro_cols = ['HEADACHE', 'ALTEREDSENSORIUM', 'SEIZURES', 'SOMNOLENCE', 'NECKRIGIDITY', 'IRRITABILITY']
            skin_cols = ['PAPULARRASH', 'PUSTULARRASH', 'MACULOPAPULARRASH', 'BULLAE']
            systemic_cols = ['MYALGIA', 'ARTHRALGIA', 'CHILLS', 'RIGORS', 'MALAISE']
            
            df['symptom_count'] = df[symptom_cols].sum(axis=1)
            
            resp_present = [c for c in respiratory_cols if c in df.columns]
            df['respiratory_symptoms'] = df[resp_present].sum(axis=1) if resp_present else 0
            
            gi_present = [c for c in gi_cols if c in df.columns]  
            df['gi_symptoms'] = df[gi_present].sum(axis=1) if gi_present else 0
            
            neuro_present = [c for c in neuro_cols if c in df.columns]
            df['neuro_symptoms'] = df[neuro_present].sum(axis=1) if neuro_present else 0
            
            skin_present = [c for c in skin_cols if c in df.columns]
            df['skin_symptoms'] = df[skin_present].sum(axis=1) if skin_present else 0
            
            systemic_present = [c for c in systemic_cols if c in df.columns]
            df['systemic_symptoms'] = df[systemic_present].sum(axis=1) if systemic_present else 0
            
            df['symptom_diversity'] = (df[symptom_cols] > 0).sum(axis=1)
            
            # Temporal features
            if 'month' in df.columns:
                def get_season(month):
                    if month in [12, 1, 2]: return 0
                    elif month in [3, 4, 5]: return 1
                    elif month in [6, 7, 8, 9]: return 2
                    else: return 3
                
                df['season'] = df['month'].apply(get_season)
                
                if 'ismonsoon' not in df.columns:
                    df['ismonsoon'] = df['month'].isin([6, 7, 8, 9]).astype(int)
                if 'iswinter' not in df.columns:
                    df['iswinter'] = df['month'].isin([12, 1, 2]).astype(int)
                
                if 'monthsin' not in df.columns:
                    df['monthsin'] = np.sin(2 * np.pi * df['month'] / 12)
                if 'monthcos' not in df.columns:
                    df['monthcos'] = np.cos(2 * np.pi * df['month'] / 12)
                
                df['week_of_year'] = df['month'] * 4
                df['day_of_year'] = df['month'] * 30
                df['quarter'] = ((df['month'] - 1) // 3) + 1
            
            # District/State encoding
            if 'districtencoded' in df.columns and 'district_encoded' not in df.columns:
                df['district_encoded'] = df['districtencoded']
            elif 'district_encoded' not in df.columns:
                df['district_encoded'] = 0
                
            if 'labstate' in df.columns and 'lab_state' not in df.columns:
                df['lab_state'] = df['labstate']
            elif 'lab_state' not in df.columns:
                df['lab_state'] = df.get('labstate', 0)
            
            if 'year' in df.columns:
                df['year_normalized'] = (df['year'] - 2012) / (2026 - 2012)
            else:
                df['year_normalized'] = 0.5
            
            # Interaction features
            if 'season' in df.columns:
                df['monsoon_respiratory'] = df.get('ismonsoon', 0) * df['respiratory_symptoms']
                df['winter_respiratory'] = df.get('iswinter', 0) * df['respiratory_symptoms']
                df['monsoon_fever'] = df.get('ismonsoon', 0) * df.get('FEVER', 0)
                df['state_season'] = df['lab_state'] * 10 + df['season']
                df['district_season'] = df['district_encoded'] * 10 + df['season']
                df['district_month'] = df['district_encoded'] * 100 + df.get('month', 1)
            
            df['state_respiratory'] = df['lab_state'] * df['respiratory_symptoms']
            df['state_fever'] = df['lab_state'] * df.get('FEVER', 0)
            df['state_gi'] = df['lab_state'] * df['gi_symptoms']
            df['fever_respiratory'] = df.get('FEVER', 0) * df['respiratory_symptoms']
            df['fever_gi'] = df.get('FEVER', 0) * df['gi_symptoms']
            df['fever_neuro'] = df.get('FEVER', 0) * df['neuro_symptoms'] 
            df['fever_skin'] = df.get('FEVER', 0) * df['skin_symptoms']
            df['fever_duration'] = df.get('FEVER', 0) * df['durationofillness']
            df['fever_headache'] = df.get('FEVER', 0) * df.get('HEADACHE', 0)
            df['fever_cough'] = df.get('FEVER', 0) * df.get('COUGH', 0)
            df['severity_score'] = df['symptom_count'] * df['durationofillness']
            df['age_symptom'] = df['age'] * df['symptom_count']
            df['age_duration'] = df['age'] * df['durationofillness']
            df['patienttype_age'] = df.get('PATIENTTYPE', 1) * df['age_group']
            df['sex_respiratory'] = df.get('SEX', 1) * df['respiratory_symptoms']
            df['duration_symptom_ratio'] = df['durationofillness'] / (df['symptom_count'] + 1)
            
            df = df.replace([np.inf, -np.inf], 0).fillna(0)
            
            # ========== FEATURE VALIDATION & COMPLETION ==========
            # Ensure ALL expected features exist (add missing ones with default value 0)
            for col in cont_cols:
                if col not in df.columns:
                    df[col] = 0.0
                    logger.debug(f"Added missing continuous feature '{col}' with default value 0")
            
            for col in binary_cols:
                if col not in df.columns:
                    df[col] = 0
                    logger.debug(f"Added missing binary feature '{col}' with default value 0")
            
            for col in cat_cols:
                if col not in df.columns:
                    df[col] = 0
                    logger.debug(f"Added missing categorical feature '{col}' with default value 0")
            
            # ========== STANDARD PREPROCESSING ==========
            
            # Continuous - now all columns are guaranteed to exist
            X_cont = imp_cont.transform(df[cont_cols])
            X_cont = scaler.transform(X_cont).astype(np.float32)
            
            # Binary - now all columns are guaranteed to exist
            X_bin = imp_bin.transform(df[binary_cols]).astype(np.float32)
            
            # Categorical - now all columns are guaranteed to exist
            X_cat_list = []
            for col in cat_cols:
                le = le_dict[col]
                val = str(df[col].values[0])
                mapping = dict(zip(le.classes_, range(len(le.classes_))))
                encoded_val = mapping.get(val, 0)  # 0 for unknown categories
                X_cat_list.append(encoded_val)
            
            X_cat = np.array([X_cat_list], dtype=np.int64) if cat_cols else np.zeros((1, 0), dtype=np.int64)
            
            # Convert to PyTorch tensors
            xb = torch.tensor(X_bin, dtype=torch.float32).to(DEVICE)
            xc = torch.tensor(X_cont, dtype=torch.float32).to(DEVICE)
            xcat = torch.tensor(X_cat, dtype=torch.long).to(DEVICE)
            
            return xb, xc, xcat
            
        except Exception as e:
            logger.error(f"Preprocessing error: {e}")
            raise
    
    def predict(self, patient_data):
        """Complete prediction workflow"""
        if self.model1 is None or self.model2 is None:
            raise RuntimeError("Models not loaded")
        
        try:
            # Preprocess for Model 1
            xb1, xc1, xcat1 = self.preprocess_features(patient_data, self.preprocessing1)
            
            # Model 1 prediction
            with torch.no_grad():
                logits1 = self.model1(xb1, xc1, xcat1)
                y_pred_proba = torch.softmax(logits1, dim=1)[0].cpu().numpy()
            
            y_pred = np.argmax(y_pred_proba)
            top_5_indices = np.argsort(y_pred_proba)[-5:][::-1]
            
            # Check for "Other_Viruses"
            second_model_results = None
            if 15 in top_5_indices:
                xb2, xc2, xcat2 = self.preprocess_features(patient_data, self.preprocessing2)
                
                with torch.no_grad():
                    logits2 = self.model2(xb2, xc2, xcat2)
                    y_pred_proba_m2 = torch.softmax(logits2, dim=1)[0].cpu().numpy()
                
                y_pred_m2 = np.argmax(y_pred_proba_m2)
                top_5_indices_m2 = np.argsort(y_pred_proba_m2)[-5:][::-1]
                
                second_model_results = {
                    'prediction': y_pred_m2,
                    'probabilities': y_pred_proba_m2,
                    'top_5': top_5_indices_m2
                }
            
            return {
                'y_pred': y_pred,
                'y_pred_proba': y_pred_proba,
                'top_5_indices': top_5_indices,
                'second_model_results': second_model_results
            }
            
        except Exception as e:
            logger.error(f"Prediction error: {e}")
            raise


# ============================================================================
# SINGLETON PREDICTOR (for FastAPI caching)
# ============================================================================

_cached_predictor = None

def get_virus_predictor():
    """Get or create a cached VirusPredictor instance"""
    global _cached_predictor
    if _cached_predictor is None:
        _cached_predictor = VirusPredictor()
    return _cached_predictor