File size: 10,608 Bytes
9686dbe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
MamaGuard β€” Data Pipeline
Loads UCI CSV, cleans, scales, builds sequences, and splits into train/val.
"""

import pandas as pd
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split

# ─── Constants ────────────────────────────────────────────────────────────────

FEATURE_COLS = ['Age', 'SystolicBP', 'DiastolicBP', 'BS', 'BodyTemp', 'HeartRate']
LABEL_COL    = 'RiskLevel'
LABEL_MAP    = {'low risk': 0, 'mid risk': 1, 'high risk': 2}
SEQ_LEN      = 5


# ─── Data quality scorer ─────────────────────────────────────────────────────

def compute_data_quality(row: pd.Series) -> float:
    """Returns 0.0–1.0 indicating how complete and plausible the data is."""
    score, total = 0.0, 0

    for col in FEATURE_COLS:
        total += 1
        if pd.notna(row.get(col)):
            score += 1.0

    plausibility = {
        'Age': (10, 60), 'SystolicBP': (70, 200), 'DiastolicBP': (40, 130),
        'BS': (3.0, 20.0), 'BodyTemp': (35.0, 42.0), 'HeartRate': (40, 160),
    }
    for col, (lo, hi) in plausibility.items():
        total += 1
        val = row.get(col)
        if pd.notna(val) and lo <= val <= hi:
            score += 1.0

    return round(score / total, 3)


# ─── Sequence builders ────────────────────────────────────────────────────────

def build_type_a_sequences(X: np.ndarray, y: np.ndarray,
                            quality: np.ndarray) -> tuple:
    """Type A β€” Within-risk-group sliding window sequences."""
    X_seq, y_seq, q_seq = [], [], []

    for risk_label in [0, 1, 2]:
        idx = np.where(y == risk_label)[0]
        if len(idx) < SEQ_LEN:
            continue

        age_order  = np.argsort(X[idx, 0])
        idx_sorted = idx[age_order]

        for i in range(len(idx_sorted) - SEQ_LEN + 1):
            seq_idx = idx_sorted[i : i + SEQ_LEN]
            X_seq.append(X[seq_idx].copy())
            y_seq.append(risk_label)
            q_seq.append(quality[seq_idx].mean())

    return X_seq, y_seq, q_seq


def build_type_b_sequences(X: np.ndarray, y: np.ndarray,
                            rng: np.random.Generator) -> tuple:
    """Type B β€” Synthetic escalation sequences for mid and high risk."""
    X_seq, y_seq, q_seq = [], [], []

    for risk_label in [1, 2]:
        indices = np.where(y == risk_label)[0]
        for idx in indices:
            anchor = X[idx]
            seq = []
            for step in range(SEQ_LEN):
                fraction = 0.30 + 0.70 * (step / (SEQ_LEN - 1))
                noise = rng.normal(0, 0.05, anchor.shape)
                seq.append(anchor * fraction + noise)
            X_seq.append(np.array(seq, dtype=np.float32))
            y_seq.append(risk_label)
            q_seq.append(1.0)

    return X_seq, y_seq, q_seq


def build_type_c_sequences(X: np.ndarray, y: np.ndarray,
                            rng: np.random.Generator,
                            n_per_transition: int = 80) -> tuple:
    """Type C β€” Cross-risk deterioration sequences (low->mid, low->high, mid->high)."""
    X_seq, y_seq, q_seq = [], [], []

    low_idx  = np.where(y == 0)[0]
    mid_idx  = np.where(y == 1)[0]
    high_idx = np.where(y == 2)[0]

    if len(low_idx) == 0 or len(mid_idx) == 0 or len(high_idx) == 0:
        return X_seq, y_seq, q_seq

    def blend(a: np.ndarray, b: np.ndarray, frac: float = 0.5) -> np.ndarray:
        """Linear interpolation between two rows + small noise."""
        return a * (1 - frac) + b * frac + rng.normal(0, 0.05, a.shape)

    def pick(indices: np.ndarray) -> np.ndarray:
        """Pick one random row from an index array."""
        return X[rng.choice(indices)]

    # Transition 1: LOW -> HIGH
    for _ in range(n_per_transition):
        seq = [
            pick(low_idx), pick(low_idx),
            blend(pick(low_idx), pick(high_idx)),
            pick(high_idx), pick(high_idx),
        ]
        X_seq.append(np.array(seq, dtype=np.float32))
        y_seq.append(2)
        q_seq.append(1.0)

    # Transition 2: LOW -> MID
    for _ in range(n_per_transition):
        seq = [
            pick(low_idx), pick(low_idx), pick(low_idx),
            pick(mid_idx), pick(mid_idx),
        ]
        X_seq.append(np.array(seq, dtype=np.float32))
        y_seq.append(1)
        q_seq.append(1.0)

    # Transition 3: MID -> HIGH
    for _ in range(n_per_transition):
        seq = [
            pick(mid_idx), pick(mid_idx),
            blend(pick(mid_idx), pick(high_idx)),
            pick(high_idx), pick(high_idx),
        ]
        X_seq.append(np.array(seq, dtype=np.float32))
        y_seq.append(2)
        q_seq.append(1.0)

    return X_seq, y_seq, q_seq


# ─── Internal helper ─────────────────────────────────────────────────────────

def _build_all_sequences(X: np.ndarray, y: np.ndarray,
                          quality: np.ndarray,
                          rng: np.random.Generator,
                          include_augmentation: bool,
                          label: str = "") -> tuple:
    """Builds all sequence types from a given set of scaled rows."""
    X_all, y_all, q_all = [], [], []

    a_X, a_y, a_q = build_type_a_sequences(X, y, quality)
    X_all += a_X; y_all += a_y; q_all += a_q
    print(f"  {label} Type A (within-class):  {len(a_X):5d} sequences")

    if include_augmentation:
        b_X, b_y, b_q = build_type_b_sequences(X, y, rng)
        X_all += b_X; y_all += b_y; q_all += b_q
        print(f"  {label} Type B (escalation):    {len(b_X):5d} sequences")

        min_cls = min(sum(y == 0), sum(y == 1), sum(y == 2))
        if min_cls >= SEQ_LEN:
            c_X, c_y, c_q = build_type_c_sequences(X, y, rng)
            X_all += c_X; y_all += c_y; q_all += c_q
            print(f"  {label} Type C (cross-risk):    {len(c_X):5d} sequences")
        else:
            print(f"  {label} Type C skipped (smallest class has {min_cls} rows)")

    return X_all, y_all, q_all


# ─── Main pipeline ────────────────────────────────────────────────────────────

def load_and_preprocess(csv_path: str):
    """
    Returns pre-split, pre-scaled sequence arrays + the fitted scaler.

    Returns:
        X_train, y_train, q_train, X_val, y_val, q_val, scaler
    """

    # 1. Load
    df = pd.read_csv(csv_path)
    print(f"Loaded {len(df)} rows from {csv_path}")
    df[LABEL_COL] = df[LABEL_COL].str.lower().str.strip()

    # 2. Quality scores BEFORE imputation
    quality_per_row = df.apply(compute_data_quality, axis=1).values

    # 3. Impute missing values
    for col in FEATURE_COLS:
        df[col] = df[col].fillna(df[col].median())

    X_raw = df[FEATURE_COLS].values.astype(np.float32)
    y_raw = df[LABEL_COL].map(LABEL_MAP).values

    # 4. Split raw rows first (prevents data leakage and sequence overlap)
    (X_train_raw, X_val_raw,
     y_train_raw, y_val_raw,
     q_train_raw, q_val_raw) = train_test_split(
        X_raw, y_raw, quality_per_row,
        test_size=0.2, random_state=42, stratify=y_raw
    )
    print(f"Raw split -> train: {len(X_train_raw)} rows | val: {len(X_val_raw)} rows")
    print(f"Train classes: low={sum(y_train_raw==0)} "
          f"mid={sum(y_train_raw==1)} high={sum(y_train_raw==2)}")
    print(f"Val   classes: low={sum(y_val_raw==0)} "
          f"mid={sum(y_val_raw==1)} high={sum(y_val_raw==2)}")

    # 5. Fit scaler on train rows only
    scaler         = StandardScaler()
    X_train_scaled = scaler.fit_transform(X_train_raw)
    X_val_scaled   = scaler.transform(X_val_raw)

    # 6. Build sequences separately
    rng = np.random.default_rng(42)

    print("\nBuilding TRAINING sequences:")
    train_seqs = _build_all_sequences(
        X_train_scaled, y_train_raw, q_train_raw,
        rng, include_augmentation=True, label="Train"
    )

    print("\nBuilding VALIDATION sequences:")
    val_seqs = _build_all_sequences(
        X_val_scaled, y_val_raw, q_val_raw,
        rng, include_augmentation=False, label="Val"
    )

    X_train, y_train, q_train = [np.array(a) for a in train_seqs]
    X_val,   y_val,   q_val   = [np.array(a) for a in val_seqs]

    print(f"\nFinal -> Train: {len(X_train)} seqs | Val: {len(X_val)} seqs")
    print(f"Train dist: {dict(zip(*np.unique(y_train, return_counts=True)))}")
    print(f"Val   dist: {dict(zip(*np.unique(y_val,   return_counts=True)))}")

    return (X_train.astype(np.float32), y_train.astype(np.int64),
            q_train.astype(np.float32),
            X_val.astype(np.float32),   y_val.astype(np.int64),
            q_val.astype(np.float32),
            scaler)


# ─── PyTorch Dataset ──────────────────────────────────────────────────────────

class MaternalDataset(Dataset):
    def __init__(self, X, y, quality):
        self.X = torch.tensor(np.array(X), dtype=torch.float32)
        self.y = torch.tensor(np.array(y), dtype=torch.long)
        self.q = torch.tensor(np.array(quality), dtype=torch.float32)

    def __len__(self):
        return len(self.X)

    def __getitem__(self, idx):
        return self.X[idx], self.y[idx], self.q[idx]


def get_dataloaders(csv_path: str, batch_size: int = 32):
    """Returns train/val DataLoaders and the fitted scaler."""
    (X_train, y_train, q_train,
     X_val,   y_val,   q_val,
     scaler) = load_and_preprocess(csv_path)

    train_ds = MaternalDataset(X_train, y_train, q_train)
    val_ds   = MaternalDataset(X_val,   y_val,   q_val)

    train_loader = DataLoader(train_ds, batch_size=batch_size, shuffle=True)
    val_loader   = DataLoader(val_ds,   batch_size=batch_size, shuffle=False)

    print(f"\nDataLoaders ready:")
    print(f"  Train: {len(train_ds)} sequences | "
          f"{len(train_loader)} batches of {batch_size}")
    print(f"  Val:   {len(val_ds)} sequences | "
          f"{len(val_loader)} batches")

    return train_loader, val_loader, scaler