File size: 12,734 Bytes
a08a4d0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
encode.py – Traffic data encoder for LSTM traffic flow prediction

This module provides TrafficDataEncoder for processing 5-minute traffic sensor data
into sequences suitable for LSTM training. Key features:
- Sensor-safe windowing (no cross-sensor leakage)
- Feature engineering (time, geographic, categorical)
- Speed-based class weighting support
- Robust missing value handling
"""

from __future__ import annotations

import numpy as np
import pandas as pd
from sklearn.preprocessing import OrdinalEncoder, StandardScaler
from sklearn.utils.validation import check_is_fitted
from typing import List, Tuple, Dict, Optional
import joblib
from pathlib import Path


class TrafficDataEncoder:
    """
    Encodes traffic sensor data into sequences for LSTM training.
    
    Features:
    - Geographic coordinates (lat/lon -> x/y km)
    - Time features (hour, day of week)
    - Categorical encoding (direction, weather)
    - Speed-based class weighting
    - Sensor-safe windowing
    """
    
    def __init__(
        self,
        seq_len: int = 12,  # 12 * 5min = 1 hour history
        horizon: int = 1,   # predict 1 step ahead (5 minutes)
        target_col: str = "speed_mph"
    ):
        self.seq_len = seq_len
        self.horizon = horizon
        self.target_col = target_col
        
        # Feature columns
        self.cat_cols = ["direction", "weather"]
        self.num_cols = [
            "lanes", "% Observed", "Latitude", "Longitude",
            "hour_sin", "hour_cos", "dow_sin", "dow_cos"
        ]
        
        # Fitted components
        self.ordinal_encoder: Optional[OrdinalEncoder] = None
        self.scaler: Optional[StandardScaler] = None
        self.num_medians: Dict[str, float] = {}
        self.is_fitted = False
    
    def _ensure_sensor_id_and_sort(self, df: pd.DataFrame) -> pd.DataFrame:
        """Create sensor_id and sort by sensor and time."""
        df = df.copy()
        
        # Create sensor_id from coordinates
        if "sensor_id" not in df.columns:
            df["sensor_id"] = (
                df["Latitude"].round(6).astype(str) + ";" + 
                df["Longitude"].round(6).astype(str)
            )
        
        # Parse time and sort
        df["Time"] = pd.to_datetime(df["Time"], errors="coerce")
        return df.sort_values(["sensor_id", "Time"]).reset_index(drop=True)
    
    def _add_time_features(self, df: pd.DataFrame) -> pd.DataFrame:
        """Add cyclical time features."""
        dt = pd.to_datetime(df["Time"], errors="coerce")
        hour = dt.dt.hour + dt.dt.minute / 60.0
        dow = dt.dt.dayofweek
        
        df["hour_sin"] = np.sin(2 * np.pi * hour / 24)
        df["hour_cos"] = np.cos(2 * np.pi * hour / 24)
        df["dow_sin"] = np.sin(2 * np.pi * dow / 7)
        df["dow_cos"] = np.cos(2 * np.pi * dow / 7)
        
        return df
    
    def _clean_numeric(self, df: pd.DataFrame) -> pd.DataFrame:
        """Clean and convert numeric columns."""
        # Ensure lanes is numeric
        df["lanes"] = pd.to_numeric(df.get("lanes", 0), errors="coerce")
        
        # Ensure % Observed is numeric
        df["% Observed"] = pd.to_numeric(df.get("% Observed", 100), errors="coerce")
        
        return df
    
    def _compute_speed_weights(self, y: np.ndarray) -> Dict[str, float]:
        """Compute class weights for speed-based weighting."""
        # Define speed classes based on user's experience
        low_mask = y <= 30
        high_mask = y >= 60
        medium_mask = ~(low_mask | high_mask)
        
        n_low = low_mask.sum()
        n_medium = medium_mask.sum()
        n_high = high_mask.sum()
        n_total = len(y)
        
        print(f"Speed distribution:")
        print(f"  Low (≤30): {n_low} samples ({n_low/n_total*100:.1f}%)")
        print(f"  Medium (30-60): {n_medium} samples ({n_medium/n_total*100:.1f}%)")
        print(f"  High (≥60): {n_high} samples ({n_high/n_total*100:.1f}%)")
        
        # Compute inverse frequency weights
        if n_low > 0 and n_medium > 0 and n_high > 0:
            weight_low = n_total / (3 * n_low)
            weight_medium = n_total / (3 * n_medium)
            weight_high = n_total / (3 * n_high)
        else:
            weight_low = weight_medium = weight_high = 1.0
        
        print(f"Class weights: Low={weight_low:.2f}, Medium={weight_medium:.2f}, High={weight_high:.2f}")
        
        return {
            "weight_low": weight_low,
            "weight_medium": weight_medium,
            "weight_high": weight_high,
            "low_threshold": 30,
            "high_threshold": 60
        }
    
    def fit(self, df: pd.DataFrame) -> "TrafficDataEncoder":
        """Fit the encoder on training data."""
        print("Fitting encoder...")
        
        # Preprocess data
        df = self._ensure_sensor_id_and_sort(df)
        df = self._add_time_features(df)
        df = self._clean_numeric(df)
        
        # Handle missing values
        df[self.cat_cols] = df[self.cat_cols].fillna("UNK")
        self.num_medians = df[self.num_cols].median(numeric_only=True).to_dict()
        df[self.num_cols] = df[self.num_cols].fillna(self.num_medians)
        
        # Fit encoders
        self.ordinal_encoder = OrdinalEncoder(
            handle_unknown="use_encoded_value", 
            unknown_value=-1
        )
        self.ordinal_encoder.fit(df[self.cat_cols])
        
        self.scaler = StandardScaler()
        self.scaler.fit(df[self.num_cols])
        
        self.is_fitted = True
        print("Encoder fitted successfully")
        return self
    
    def _preprocess(self, df: pd.DataFrame) -> pd.DataFrame:
        """Apply preprocessing steps."""
        df = self._ensure_sensor_id_and_sort(df)
        df = self._add_time_features(df)
        df = self._clean_numeric(df)
        
        # Handle missing values using fitted medians
        df[self.cat_cols] = df[self.cat_cols].fillna("UNK")
        df[self.num_cols] = df[self.num_cols].fillna(self.num_medians)
        
        return df
    
    def transform(self, df: pd.DataFrame) -> Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
        """
        Transform data into sequences.
        
        Returns:
            X: (N, seq_len, n_features) - input sequences
            y: (N, horizon) - target values
            target_indices: (N,) - indices of target rows in original df
            timestamps: (N,) - timestamps of target rows
        """
        check_is_fitted(self, ["ordinal_encoder", "scaler", "num_medians"])
        
        df = self._preprocess(df)
        
        X_chunks = []
        y_chunks = []
        target_indices = []
        timestamps = []
        
        # Process each sensor separately to avoid cross-sensor leakage
        for sensor_id, group in df.groupby("sensor_id", sort=False):
            if len(group) < self.seq_len + self.horizon:
                continue  # Not enough data for this sensor
            
            # Encode features
            cat_features = self.ordinal_encoder.transform(group[self.cat_cols]).astype(np.float32)
            num_features = self.scaler.transform(group[self.num_cols]).astype(np.float32)
            features = np.concatenate([num_features, cat_features], axis=1)
            
            # Get targets
            targets = group[self.target_col].to_numpy(dtype=np.float32)
            group_timestamps = group["Time"].to_numpy()
            group_indices = group.index.to_numpy()
            
            # Create sliding windows
            n_windows = len(group) - self.seq_len - self.horizon + 1
            for i in range(n_windows):
                X_chunks.append(features[i:i + self.seq_len])
                y_chunks.append(targets[i + self.seq_len:i + self.seq_len + self.horizon])
                target_indices.append(group_indices[i + self.seq_len + self.horizon - 1])
                timestamps.append(group_timestamps[i + self.seq_len + self.horizon - 1])
        
        if not X_chunks:
            # Return empty arrays with correct shapes
            n_features = len(self.num_cols) + len(self.cat_cols)
            return (
                np.empty((0, self.seq_len, n_features), dtype=np.float32),
                np.empty((0, self.horizon), dtype=np.float32),
                np.empty(0, dtype=int),
                np.empty(0, dtype=object)
            )
        
        X = np.stack(X_chunks, axis=0)
        y = np.stack(y_chunks, axis=0)
        target_indices = np.array(target_indices, dtype=int)
        timestamps = np.array(timestamps)
        
        print(f"Created {len(X)} sequences from {len(df.groupby('sensor_id'))} sensors")
        return X, y, target_indices, timestamps
    
    def fit_transform(self, df: pd.DataFrame) -> Tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
        """Fit encoder and transform data in one step."""
        return self.fit(df).transform(df)
    
    def get_speed_weights(self, y: np.ndarray) -> Dict[str, float]:
        """Get speed-based class weights for weighted loss."""
        return self._compute_speed_weights(y)
    
    def save(self, filepath: str) -> None:
        """Save the fitted encoder."""
        if not self.is_fitted:
            raise ValueError("Encoder must be fitted before saving")
        
        joblib.dump(self, filepath)
        print(f"Encoder saved to {filepath}")
    
    @classmethod
    def load(cls, filepath: str) -> "TrafficDataEncoder":
        """Load a fitted encoder."""

        try:
            encoder = joblib.load(filepath)
            if not isinstance(encoder, cls):
                raise ValueError(f"Loaded object is not a {cls.__name__}")
            return encoder
        except AttributeError as e:
            if "TrafficDataEncoder" in str(e):
                # Handle the case where encoder was saved from a different module context
                print("Warning: Encoder was saved from different module context. Reconstructing...")
                
                # Use a more robust approach with joblib
                import sys
                import types
                
                # Temporarily modify sys.modules to include our class
                original_main = sys.modules.get('__main__')
                temp_module = types.ModuleType('temp_encode')
                temp_module.TrafficDataEncoder = cls
                sys.modules['__main__'] = temp_module
                
                try:
                    # Now try loading with the modified module context
                    encoder = joblib.load(filepath)
                    if not isinstance(encoder, cls):
                        raise ValueError(f"Loaded object is not a {cls.__name__}")
                    return encoder
                finally:
                    # Restore original __main__ module
                    if original_main is not None:
                        sys.modules['__main__'] = original_main
                    else:
                        del sys.modules['__main__']
            else:
                raise e


def main():
    """CLI interface for encoding data."""
    import argparse
    
    parser = argparse.ArgumentParser(description="Encode traffic data for LSTM training")
    parser.add_argument("csv_file", help="Path to CSV file with traffic data")
    parser.add_argument("--seq_len", type=int, default=12, help="Sequence length (default: 12)")
    parser.add_argument("--horizon", type=int, default=1, help="Prediction horizon (default: 1)")
    parser.add_argument("--target_col", default="speed_mph", help="Target column name")
    parser.add_argument("--save_encoder", help="Path to save fitted encoder")
    parser.add_argument("--output", help="Path to save encoded data (optional)")
    
    args = parser.parse_args()
    
    # Load data
    print(f"Loading data from {args.csv_file}")
    df = pd.read_csv(args.csv_file)
    
    # Create and fit encoder
    encoder = TrafficDataEncoder(
        seq_len=args.seq_len,
        horizon=args.horizon,
        target_col=args.target_col
    )
    
    X, y, target_indices, timestamps = encoder.fit_transform(df)
    
    print(f"Encoded data shapes:")
    print(f"  X: {X.shape}")
    print(f"  y: {y.shape}")
    print(f"  Target indices: {len(target_indices)}")
    print(f"  Timestamps: {len(timestamps)}")
    
    # Save encoder if requested
    if args.save_encoder:
        encoder.save(args.save_encoder)
    
    # Save encoded data if requested
    if args.output:
        np.savez(args.output, X=X, y=y, target_indices=target_indices, timestamps=timestamps)
        print(f"Encoded data saved to {args.output}")


if __name__ == "__main__":
    main()