Spaces:
Sleeping
Sleeping
File size: 20,037 Bytes
5ccd893 | 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 | """
Feature Engineering Model
Handles weather feature engineering computations with proper NaN handling
v3 FIX: Uses training-dataset global statistics for feature normalization
instead of per-window stats (which inflated disaster features for
normal weather, causing the model to always predict 'Disaster').
"""
import numpy as np
import pandas as pd
import json
import os
from typing import Dict, List, Optional, Any, Tuple
from datetime import datetime
import logging
class WeatherFeatureModel:
"""Model for weather feature engineering operations"""
# Path to training-dataset global statistics
# These stats were computed across the ENTIRE training dataset (~123k rows × 60 days)
# and MUST be used for feature engineering to match the training distribution.
# Without them, per-window stats inflate disaster-related features for normal weather.
TRAINING_STATS_PATH = os.path.join(
os.path.dirname(os.path.abspath(__file__)),
'hazardguard', 'shared', 'training_global_stats.json'
)
# Cached training stats (loaded once)
_training_stats = None
# Original 17 weather fields from NASA POWER
WEATHER_FIELDS = [
'temperature_C',
'humidity_perc',
'wind_speed_mps',
'precipitation_mm',
'surface_pressure_hPa',
'solar_radiation_wm2',
'temperature_max_C',
'temperature_min_C',
'specific_humidity_g_kg',
'dew_point_C',
'wind_speed_10m_mps',
'cloud_amount_perc',
'sea_level_pressure_hPa',
'surface_soil_wetness_perc',
'wind_direction_10m_degrees',
'evapotranspiration_wm2',
'root_zone_soil_moisture_perc'
]
# 19 engineered features (excluding precip_intensity_mm_day as requested)
ENGINEERED_FEATURES = [
'temp_normalized',
'temp_range',
'discomfort_index',
'heat_index',
'wind_precip_interaction',
'solar_temp_ratio',
'pressure_anomaly',
'high_precip_flag',
'adjusted_humidity',
'wind_chill',
'solar_radiation_anomaly',
'weather_severity_score',
'moisture_stress_index',
'evaporation_deficit',
'soil_saturation_index',
'atmospheric_instability',
'drought_indicator',
'flood_risk_score',
'storm_intensity_index'
]
# Feature descriptions for documentation
FEATURE_DESCRIPTIONS = {
'temp_normalized': 'Temperature normalized between min/max',
'temp_range': 'Diurnal temperature range (max - min)',
'discomfort_index': 'Temperature-Humidity Index (THI)',
'heat_index': 'Apparent temperature accounting for humidity',
'wind_precip_interaction': 'Wind speed × precipitation interaction',
'solar_temp_ratio': 'Solar radiation to temperature ratio',
'pressure_anomaly': 'Deviation from mean surface pressure',
'high_precip_flag': 'Binary flag for precipitation >50mm',
'adjusted_humidity': 'Humidity adjusted for temperature',
'wind_chill': 'Perceived temperature with wind effect',
'solar_radiation_anomaly': 'Deviation from mean solar radiation',
'weather_severity_score': 'Composite severity index (0-1)',
'moisture_stress_index': 'Evaporation vs precipitation stress',
'evaporation_deficit': 'Deviation from mean evapotranspiration',
'soil_saturation_index': 'Combined surface + root zone moisture',
'atmospheric_instability': 'Pressure difference + temperature range',
'drought_indicator': 'Low precip + high temp + low soil moisture',
'flood_risk_score': 'High precip + saturated soil + low evap',
'storm_intensity_index': 'Wind + precipitation + pressure drop'
}
@classmethod
def validate_weather_data(cls, weather_data: Dict[str, List]) -> Dict[str, Any]:
"""
Validate weather data for feature engineering
Args:
weather_data: Dictionary with weather field lists
Returns:
Validation results
"""
errors = []
warnings = []
# Check required fields
missing_fields = set(cls.WEATHER_FIELDS) - set(weather_data.keys())
if missing_fields:
errors.append(f"Missing required weather fields: {missing_fields}")
# Check list lengths and data quality
days_count = None
for field, values in weather_data.items():
if not isinstance(values, list):
errors.append(f"Field {field} must be a list, got {type(values)}")
continue
if days_count is None:
days_count = len(values)
elif len(values) != days_count:
warnings.append(f"Field {field} has {len(values)} values, expected {days_count}")
# Check for all NaN lists
for field, values in weather_data.items():
if isinstance(values, list):
valid_count = sum(1 for v in values if v is not None and not np.isnan(float(v)) if v != -999)
if valid_count == 0:
warnings.append(f"Field {field} contains only NaN/missing values")
return {
'valid': len(errors) == 0,
'errors': errors,
'warnings': warnings,
'days_count': days_count,
'field_count': len(weather_data)
}
@classmethod
def load_training_stats(cls) -> Dict[str, float]:
"""
Load global statistics from the training dataset.
These stats were pre-computed across the ENTIRE training dataset
(~123k rows × 60 days = ~7.4M data points per weather variable).
Using per-window stats instead leads to feature scale mismatch:
e.g., precip_max=994mm (training) vs precip_max=10mm (single window)
which inflates disaster features 100x and causes false positives.
Returns:
Dictionary of training-dataset global statistics
"""
logger = logging.getLogger(__name__)
# Return cached stats if already loaded
if cls._training_stats is not None:
return cls._training_stats
# Try loading from JSON file
stats_path = cls.TRAINING_STATS_PATH
if os.path.exists(stats_path):
try:
with open(stats_path, 'r') as f:
loaded = json.load(f)
cls._training_stats = {
'temp_min': loaded.get('temp_min', -53.76),
'temp_max': loaded.get('temp_max', 44.18),
'temp_mean': loaded.get('temp_mean', 21.5325),
'temp_max_mean': loaded.get('temp_max_mean', 25.7065),
'pressure_mean': loaded.get('pressure_mean', 93.7966),
'sea_pressure_mean': loaded.get('sea_pressure_mean', 101.0376),
'solar_mean': loaded.get('solar_mean', 4.9125),
'precip_max': max(loaded.get('precip_max', 994.16), 1),
'wind_max': max(loaded.get('wind_max', 25.32), 1),
'evap_mean': loaded.get('evap_mean', 0.5756),
}
logger.info(f"[STATS] Loaded training global stats from {stats_path}")
logger.info(f" precip_max={cls._training_stats['precip_max']}, "
f"wind_max={cls._training_stats['wind_max']}, "
f"temp_range=[{cls._training_stats['temp_min']}, {cls._training_stats['temp_max']}]")
return cls._training_stats
except Exception as e:
logger.error(f"[STATS] Error loading training stats: {e}")
else:
logger.warning(f"[STATS] Training stats file not found: {stats_path}")
# Fallback: hardcoded values from actual training dataset computation
# (computed from VALIDATED_LAT-LONG_CLEAN_DEDUPLICATED.xlsx, 123143 rows)
logger.warning("[STATS] Using hardcoded fallback training stats")
cls._training_stats = {
'temp_min': -53.76,
'temp_max': 44.18,
'temp_mean': 21.5325,
'temp_max_mean': 25.7065,
'pressure_mean': 93.7966,
'sea_pressure_mean': 101.0376,
'solar_mean': 4.9125,
'precip_max': 994.16,
'wind_max': 25.32,
'evap_mean': 0.5756,
}
return cls._training_stats
@classmethod
def compute_global_stats(cls, weather_data: Dict[str, List]) -> Dict[str, float]:
"""
Return global statistics for feature engineering.
FIXED (v3): Now returns training-dataset stats instead of computing
per-window stats. The per-window approach inflated disaster features
for normal weather (e.g., precip/precip_max_window ≈ 1.0 vs
precip/precip_max_training ≈ 0.01) causing 100% disaster predictions.
Args:
weather_data: Weather data dictionary (ignored — uses training stats)
Returns:
Dictionary of global statistics from training dataset
"""
return cls.load_training_stats()
@classmethod
def safe_float(cls, value, default=np.nan):
"""Safely convert value to float, return default for NaN/invalid values"""
if value is None or value == -999:
return default
try:
float_val = float(value)
return float_val if not np.isnan(float_val) else default
except (ValueError, TypeError):
return default
@classmethod
def compute_engineered_features(cls, weather_data: Dict[str, List],
event_duration: float = 1.0) -> Dict[str, List[float]]:
"""
Compute all engineered features from weather data
Args:
weather_data: Dictionary with weather field lists (60 days each)
event_duration: Duration of event in days for intensity calculations
Returns:
Dictionary with engineered feature lists
"""
# Validate data
validation = cls.validate_weather_data(weather_data)
if not validation['valid']:
raise ValueError(f"Invalid weather data: {validation['errors']}")
days_count = validation['days_count']
if days_count is None or days_count == 0:
raise ValueError("No weather data provided")
# Compute global statistics
stats = cls.compute_global_stats(weather_data)
# Initialize feature lists
features = {feature: [] for feature in cls.ENGINEERED_FEATURES}
# Process each day
for day_idx in range(days_count):
# Extract daily values with safe conversion
temp = cls.safe_float(weather_data.get('temperature_C', [None] * days_count)[day_idx])
temp_max = cls.safe_float(weather_data.get('temperature_max_C', [None] * days_count)[day_idx])
temp_min = cls.safe_float(weather_data.get('temperature_min_C', [None] * days_count)[day_idx])
humidity = cls.safe_float(weather_data.get('humidity_perc', [None] * days_count)[day_idx])
spec_humidity = cls.safe_float(weather_data.get('specific_humidity_g_kg', [None] * days_count)[day_idx])
dew_point = cls.safe_float(weather_data.get('dew_point_C', [None] * days_count)[day_idx])
wind = cls.safe_float(weather_data.get('wind_speed_mps', [None] * days_count)[day_idx])
wind_10m = cls.safe_float(weather_data.get('wind_speed_10m_mps', [None] * days_count)[day_idx])
precip = cls.safe_float(weather_data.get('precipitation_mm', [None] * days_count)[day_idx])
pressure = cls.safe_float(weather_data.get('surface_pressure_hPa', [None] * days_count)[day_idx])
sea_pressure = cls.safe_float(weather_data.get('sea_level_pressure_hPa', [None] * days_count)[day_idx])
solar = cls.safe_float(weather_data.get('solar_radiation_wm2', [None] * days_count)[day_idx])
cloud = cls.safe_float(weather_data.get('cloud_amount_perc', [None] * days_count)[day_idx])
soil_wetness = cls.safe_float(weather_data.get('surface_soil_wetness_perc', [None] * days_count)[day_idx])
wind_dir = cls.safe_float(weather_data.get('wind_direction_10m_degrees', [None] * days_count)[day_idx])
evap = cls.safe_float(weather_data.get('evapotranspiration_wm2', [None] * days_count)[day_idx])
root_moisture = cls.safe_float(weather_data.get('root_zone_soil_moisture_perc', [None] * days_count)[day_idx])
# Compute engineered features with NaN handling
feature_values = cls._compute_daily_features(
temp, temp_max, temp_min, humidity, spec_humidity, dew_point,
wind, wind_10m, precip, pressure, sea_pressure, solar, cloud,
soil_wetness, wind_dir, evap, root_moisture, event_duration, stats
)
# Add computed values to feature lists
for feature_name, value in zip(cls.ENGINEERED_FEATURES, feature_values):
features[feature_name].append(value)
return features
@classmethod
def _compute_daily_features(cls, temp, temp_max, temp_min, humidity, spec_humidity,
dew_point, wind, wind_10m, precip, pressure, sea_pressure,
solar, cloud, soil_wetness, wind_dir, evap, root_moisture,
event_duration, stats):
"""Compute engineered features for a single day with proper NaN handling.
FIXED (v3): temp_normalized now uses daily T2M_MIN/T2M_MAX (matching
step7 training pipeline where loop variables shadow global ones).
drought_indicator also uses daily temp_max (same shadowing effect).
"""
def safe_calc(func, *args, default=np.nan):
"""Safely execute calculation, return NaN if any input is NaN"""
try:
if any(np.isnan(arg) if isinstance(arg, (int, float)) else False for arg in args):
return default
return func(*args)
except (ZeroDivisionError, ValueError, TypeError):
return default
# 1. Temperature Normalization
# FIXED: Uses daily T2M_MIN and T2M_MAX (not global min/max of T2M)
# In step7, the loop variables `temp_min` and `temp_max` shadow the globals,
# so temp_normalized = (T2M - T2M_MIN_daily) / (T2M_MAX_daily - T2M_MIN_daily)
temp_normalized = safe_calc(
lambda: (temp - temp_min) / (temp_max - temp_min)
if temp_max != temp_min else 0
)
# 2. Temperature Range (diurnal)
temp_range = safe_calc(lambda: temp_max - temp_min)
# 3. Discomfort Index (THI)
discomfort_index = safe_calc(
lambda: temp - 0.55 * (1 - 0.01 * humidity) * (temp - 14.5)
)
# 4. Heat Index
heat_index = safe_calc(lambda: cls._calculate_heat_index(temp, humidity))
# 5. Wind-Precipitation Interaction
wind_precip_interaction = safe_calc(lambda: wind * precip)
# 6. Solar Radiation to Temperature Ratio
solar_temp_ratio = safe_calc(
lambda: solar / (abs(temp) + 0.01) if abs(temp) + 0.01 > 1e-6 else 0
)
# 7. Pressure Anomaly (surface)
pressure_anomaly = safe_calc(lambda: pressure - stats['pressure_mean'])
# 8. High Precipitation Flag (>50mm threshold)
high_precip_flag = safe_calc(lambda: int(precip > 50))
# 9. Relative Humidity Adjusted for Temperature
adjusted_humidity = safe_calc(lambda: humidity * (1 + (temp / 100)))
# 10. Wind Chill Index
wind_chill = safe_calc(lambda: cls._calculate_wind_chill(temp, wind))
# 11. Solar Radiation Anomaly
solar_anomaly = safe_calc(lambda: solar - stats['solar_mean'])
# 12. Weather Severity Score (composite)
weather_severity = safe_calc(lambda: (
(temp_normalized if not np.isnan(temp_normalized) else 0) +
(precip / stats['precip_max'] if stats['precip_max'] != 0 else 0) +
(wind / stats['wind_max'] if stats['wind_max'] != 0 else 0) +
(cloud / 100 if not np.isnan(cloud) else 0)
) / 4)
# 13. Moisture Stress Index (evaporation vs precipitation)
moisture_stress = safe_calc(
lambda: (evap - precip) / (evap + precip + 0.01)
)
# 14. Evaporation Deficit
evap_deficit = safe_calc(lambda: evap - stats['evap_mean'])
# 15. Soil Saturation Index (combined soil moisture)
soil_saturation = safe_calc(lambda: (soil_wetness + root_moisture) / 2)
# 16. Atmospheric Instability (pressure difference + temp range)
atm_instability = safe_calc(
lambda: abs(sea_pressure - pressure) + (temp_range if not np.isnan(temp_range) else 0)
)
# 17. Drought Indicator (low precip + high temp + low soil moisture)
# FIXED: Uses daily temp_max (T2M_MAX) not global stats['temp_max']
# In step7, variable shadowing means the per-day T2M_MAX is used here
drought_indicator = safe_calc(lambda: (
(1 - precip / stats['precip_max']) *
((temp - stats['temp_mean']) / max(abs(temp_max - stats['temp_mean']), 1)) *
(1 - (soil_saturation if not np.isnan(soil_saturation) else 0) / 100)
))
# 18. Flood Risk Score (high precip + saturated soil + low evap)
flood_risk = safe_calc(lambda: (
(precip / stats['precip_max']) *
((soil_saturation if not np.isnan(soil_saturation) else 0) / 100) *
(1 - evap / max(stats['evap_mean'] * 2, 1))
))
# 19. Storm Intensity Index (wind + precip + pressure drop)
storm_intensity = safe_calc(lambda: (
(wind_10m / stats['wind_max']) +
(precip / stats['precip_max']) +
(abs(pressure_anomaly if not np.isnan(pressure_anomaly) else 0) / 50)
))
return [
temp_normalized, temp_range, discomfort_index, heat_index,
wind_precip_interaction, solar_temp_ratio, pressure_anomaly,
high_precip_flag, adjusted_humidity, wind_chill, solar_anomaly,
weather_severity, moisture_stress, evap_deficit, soil_saturation,
atm_instability, drought_indicator, flood_risk, storm_intensity
]
@staticmethod
def _calculate_heat_index(temp, humidity):
"""Calculate heat index with NaN handling"""
if np.isnan(temp) or np.isnan(humidity):
return np.nan
if temp >= 27 and humidity >= 40:
return (-8.78469475556 + 1.61139411 * temp + 2.33854883889 * humidity +
-0.14611605 * temp * humidity + -0.012308094 * temp**2 +
-0.0164248277778 * humidity**2 + 0.002211732 * temp**2 * humidity +
0.00072546 * temp * humidity**2 + -0.000003582 * temp**2 * humidity**2)
else:
return temp
@staticmethod
def _calculate_wind_chill(temp, wind):
"""Calculate wind chill with NaN handling"""
if np.isnan(temp) or np.isnan(wind):
return np.nan
if temp <= 10 and wind > 0:
return (13.12 + 0.6215 * temp - 11.37 * np.power(wind, 0.16) +
0.3965 * temp * np.power(wind, 0.16))
else:
return temp |