File size: 16,513 Bytes
c49b21b |
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 |
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
import json
import logging
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
def normalize_stock_data(df_stocks):
"""
Normalize stock data to ensure consistent format for merging.
"""
logger.info("=== NORMALIZING STOCK DATA ===")
df_stocks = df_stocks.copy()
# Normalize symbol to uppercase and strip whitespace
df_stocks['symbol'] = df_stocks['symbol'].astype(str).str.upper().str.strip()
# Ensure interval_timestamp is int64 (Unix timestamp in milliseconds)
if 'interval_timestamp' in df_stocks.columns:
# If it's already numeric, ensure it's int64
df_stocks['interval_timestamp'] = pd.to_numeric(df_stocks['interval_timestamp'], errors='coerce').astype('int64')
logger.info(f"Stock timestamp range: {df_stocks['interval_timestamp'].min()} to {df_stocks['interval_timestamp'].max()}")
logger.info(f"Stock timestamp sample: {df_stocks['interval_timestamp'].head().tolist()}")
logger.info(f"Stock symbols sample: {df_stocks['symbol'].unique()[:10].tolist()}")
logger.info(f"Stock data shape: {df_stocks.shape}")
return df_stocks
def normalize_news_data(df_news):
"""
Normalize news data to ensure consistent format for merging.
"""
logger.info("=== NORMALIZING NEWS DATA ===")
df_news = df_news.copy()
# Extract entities and create individual records
news_records = []
for idx, row in df_news.iterrows():
entities = row.get('entities', [])
# Only proceed if entities is a non-empty list or ndarray
if not isinstance(entities, (list, np.ndarray)) or len(entities) == 0:
continue
# Convert published_at to timestamp
try:
if isinstance(row['published_at'], str):
published_dt = pd.to_datetime(row['published_at'])
else:
published_dt = row['published_at']
except:
logger.warning(f"Could not parse published_at for row {idx}")
continue
# Process each entity
for entity in entities:
if not isinstance(entity, dict):
continue
# Only process equity type entities with symbols
if entity.get('type') == 'equity' and 'symbol' in entity:
symbol = str(entity['symbol']).upper().strip()
# Create 30-minute intervals (matching your stock data)
interval_dt = published_dt.floor('30min')
# Convert to Unix timestamp in milliseconds
interval_timestamp = int(interval_dt.timestamp() * 1000)
news_records.append({
'symbol': symbol,
'interval_timestamp': interval_timestamp,
'published_at': published_dt,
'sentiment_score': entity.get('sentiment_score', 0),
'match_score': entity.get('match_score', 0),
'highlights_count': len(entity.get('highlights', [])),
'news_uuid': row.get('uuid', ''),
'news_title': row.get('title', ''),
'news_source': row.get('source', ''),
'relevance_score': row.get('relevance_score', 0)
})
if not news_records:
logger.warning("No valid news records found")
return pd.DataFrame()
df_news_normalized = pd.DataFrame(news_records)
logger.info(f"Normalized news data shape: {df_news_normalized.shape}")
# Print columns that are completely null and those that aren't
null_columns = [col for col in df_news_normalized.columns if df_news_normalized[col].isnull().all()]
not_null_columns = [col for col in df_news_normalized.columns if not df_news_normalized[col].isnull().all()]
print(f"Completely null columns: {null_columns}")
print(f"Non-null columns: {not_null_columns}")
logger.info(f"News symbols sample: {df_news_normalized['symbol'].unique()[:10].tolist()}")
logger.info(f"News timestamp range: {df_news_normalized['interval_timestamp'].min()} to {df_news_normalized['interval_timestamp'].max()}")
logger.info(f"News timestamp sample: {df_news_normalized['interval_timestamp'].head().tolist()}")
return df_news_normalized
def find_nearest_timestamp_matches(df_stocks, df_news, time_tolerance_minutes=30):
"""
Find the nearest timestamp matches within a tolerance window.
This handles cases where timestamps don't align exactly.
"""
logger.info(f"=== FINDING NEAREST TIMESTAMP MATCHES (tolerance: {time_tolerance_minutes} min) ===")
if df_news.empty:
return df_stocks.assign(**{col: 0 for col in [
'news_sentiment_mean', 'news_sentiment_std', 'news_sentiment_min', 'news_sentiment_max',
'news_match_score_mean', 'news_match_score_max', 'news_highlights_count',
'news_articles_count', 'latest_news_timestamp', 'news_sentiment_range',
'news_activity_score', 'news_mentions_count'
]})
# Convert tolerance to milliseconds
tolerance_ms = time_tolerance_minutes * 60 * 1000
# Get unique combinations for efficient processing
stock_keys = df_stocks[['symbol', 'interval_timestamp']].drop_duplicates()
matched_records = []
for _, stock_row in stock_keys.iterrows():
symbol = stock_row['symbol']
stock_timestamp = stock_row['interval_timestamp']
# Find news for this symbol
symbol_news = df_news[df_news['symbol'] == symbol].copy()
if symbol_news.empty:
continue
# Calculate time differences
symbol_news['time_diff'] = abs(symbol_news['interval_timestamp'] - stock_timestamp)
# Filter within tolerance
nearby_news = symbol_news[symbol_news['time_diff'] <= tolerance_ms]
if nearby_news.empty:
continue
# Aggregate the nearby news
agg_data = {
'symbol': symbol,
'interval_timestamp': stock_timestamp,
'news_sentiment_mean': nearby_news['sentiment_score'].mean(),
'news_sentiment_std': nearby_news['sentiment_score'].std(),
'news_sentiment_min': nearby_news['sentiment_score'].min(),
'news_sentiment_max': nearby_news['sentiment_score'].max(),
'news_match_score_mean': nearby_news['match_score'].mean(),
'news_match_score_max': nearby_news['match_score'].max(),
'news_highlights_count': nearby_news['highlights_count'].sum(),
'news_articles_count': len(nearby_news),
'latest_news_timestamp': nearby_news['published_at'].max(),
'news_mentions_count': len(nearby_news)
}
# Calculate additional features
agg_data['news_sentiment_range'] = agg_data['news_sentiment_max'] - agg_data['news_sentiment_min']
agg_data['news_activity_score'] = agg_data['news_match_score_mean'] + agg_data['news_match_score_max']
# Fill NaN values
for key, value in agg_data.items():
if pd.isna(value) and key not in ['symbol', 'interval_timestamp', 'latest_news_timestamp']:
agg_data[key] = 0
matched_records.append(agg_data)
if matched_records:
df_matched_news = pd.DataFrame(matched_records)
logger.info(f"Found {len(df_matched_news)} symbol-timestamp matches")
# Merge with stock data
df_result = df_stocks.merge(
df_matched_news,
on=['symbol', 'interval_timestamp'],
how='left'
)
else:
logger.warning("No timestamp matches found within tolerance")
df_result = df_stocks.copy()
# Fill remaining NaN values for stocks without news
news_columns = [
'news_sentiment_mean', 'news_sentiment_std', 'news_sentiment_min', 'news_sentiment_max',
'news_match_score_mean', 'news_match_score_max', 'news_highlights_count',
'news_articles_count', 'news_sentiment_range', 'news_activity_score', 'news_mentions_count'
]
for col in news_columns:
if col in df_result.columns:
df_result[col] = df_result[col].fillna(0)
# Report results
if 'news_articles_count' in df_result.columns:
stocks_with_news = len(df_result[df_result['news_articles_count'] > 0])
total_news_articles = df_result['news_articles_count'].sum()
logger.info(f"Successfully matched news for {stocks_with_news} stock records out of {len(df_result)}")
logger.info(f"Total news articles matched: {total_news_articles}")
return df_result
def diagnose_data_alignment(df_stocks, df_news):
"""
Diagnose alignment issues between stock and news data.
"""
logger.info("=== DATA ALIGNMENT DIAGNOSIS ===")
# Check symbol overlap
stock_symbols = set(df_stocks['symbol'].unique()) if 'symbol' in df_stocks.columns else set()
news_symbols = set(df_news['symbol'].unique()) if len(df_news) > 0 and 'symbol' in df_news.columns else set()
common_symbols = stock_symbols.intersection(news_symbols)
logger.info(f"Stock symbols: {len(stock_symbols)} unique")
logger.info(f"News symbols: {len(news_symbols)} unique")
logger.info(f"Common symbols: {len(common_symbols)}")
logger.info(f"Common symbols sample: {list(common_symbols)[:10]}")
# Check timestamp ranges
if 'interval_timestamp' in df_stocks.columns:
stock_ts_min = df_stocks['interval_timestamp'].min()
stock_ts_max = df_stocks['interval_timestamp'].max()
stock_ts_range = pd.to_datetime([stock_ts_min, stock_ts_max], unit='ms')
logger.info(f"Stock timestamp range: {stock_ts_range[0]} to {stock_ts_range[1]}")
if len(df_news) > 0 and 'interval_timestamp' in df_news.columns:
news_ts_min = df_news['interval_timestamp'].min()
news_ts_max = df_news['interval_timestamp'].max()
news_ts_range = pd.to_datetime([news_ts_min, news_ts_max], unit='ms')
logger.info(f"News timestamp range: {news_ts_range[0]} to {news_ts_range[1]}")
# Check for timestamp overlap
if 'interval_timestamp' in df_stocks.columns:
overlap_start = max(stock_ts_min, news_ts_min)
overlap_end = min(stock_ts_max, news_ts_max)
if overlap_start <= overlap_end:
overlap_range = pd.to_datetime([overlap_start, overlap_end], unit='ms')
logger.info(f"Timestamp overlap: {overlap_range[0]} to {overlap_range[1]}")
else:
logger.warning("No timestamp overlap between stock and news data")
def parse_json_news_file(news_file_path):
"""
Parse news file that contains JSON records (one per line or structured).
"""
logger.info(f"Parsing news file: {news_file_path}")
try:
# Try reading as parquet first
df_news = pd.read_parquet(news_file_path)
logger.info(f"Successfully read parquet file with shape: {df_news.shape}")
# Check if the data contains JSON strings that need parsing
if len(df_news.columns) == 1 and df_news.iloc[0, 0] and isinstance(df_news.iloc[0, 0], str):
logger.info("Detected JSON strings in single column, parsing...")
json_records = []
for idx, row in df_news.iterrows():
try:
json_data = json.loads(row.iloc[0])
json_records.append(json_data)
except json.JSONDecodeError as e:
logger.warning(f"Failed to parse JSON at row {idx}: {e}")
continue
if json_records:
df_news = pd.DataFrame(json_records)
logger.info(f"Parsed {len(json_records)} JSON records")
return df_news
except Exception as e:
logger.error(f"Error reading news file: {e}")
return pd.DataFrame()
def main(stocks_file_path, news_file_path, output_file_path, time_tolerance_minutes=30):
"""
Main function to normalize and merge stock and news data.
"""
try:
logger.info("=== STARTING DATA NORMALIZATION AND MERGE ===")
# Step 1: Load stock data
logger.info("Step 1: Loading stock data...")
df_stocks = pd.read_parquet(stocks_file_path)
logger.info(f"Loaded stock data with shape: {df_stocks.shape}")
# Step 2: Load and parse news data
logger.info("Step 2: Loading news data...")
df_news_raw = parse_json_news_file(news_file_path)
if df_news_raw.empty:
logger.warning("No news data found, creating stock data with empty news columns")
df_stocks = normalize_stock_data(df_stocks)
# Add empty news columns
for col in ['news_sentiment_mean', 'news_sentiment_std', 'news_sentiment_min',
'news_sentiment_max', 'news_match_score_mean', 'news_match_score_max',
'news_highlights_count', 'news_articles_count', 'latest_news_timestamp',
'news_sentiment_range', 'news_activity_score', 'news_mentions_count']:
df_stocks[col] = 0 if col != 'latest_news_timestamp' else None
df_stocks.to_parquet(output_file_path, index=False)
logger.info("Saved stock data with empty news columns")
return df_stocks
# Step 3: Normalize both datasets
logger.info("Step 3: Normalizing stock data...")
df_stocks_norm = normalize_stock_data(df_stocks)
logger.info("Step 4: Normalizing news data...")
df_news_norm = normalize_news_data(df_news_raw)
# Step 5: Diagnose alignment
logger.info("Step 5: Diagnosing data alignment...")
diagnose_data_alignment(df_stocks_norm, df_news_norm)
# Step 6: Find nearest timestamp matches and merge
logger.info("Step 6: Finding nearest timestamp matches and merging...")
df_merged = find_nearest_timestamp_matches(
df_stocks_norm,
df_news_norm,
time_tolerance_minutes=time_tolerance_minutes
)
# Step 7: Save results
logger.info("Step 7: Saving merged data...")
df_merged.to_parquet(output_file_path, index=False)
logger.info(f"Saved merged data to {output_file_path}")
# Final report
logger.info("=== MERGE COMPLETED ===")
logger.info(f"Final dataset shape: {df_merged.shape}")
news_cols = [col for col in df_merged.columns if col.startswith('news_')]
logger.info(f"News columns added: {len(news_cols)}")
if 'news_articles_count' in df_merged.columns:
total_articles = df_merged['news_articles_count'].sum()
records_with_news = len(df_merged[df_merged['news_articles_count'] > 0])
logger.info(f"Total news articles merged: {total_articles}")
logger.info(f"Stock records with news: {records_with_news} / {len(df_merged)}")
return df_merged
except Exception as e:
logger.error(f"Error in main process: {e}")
import traceback
logger.error(traceback.format_exc())
raise
# Example usage
if __name__ == "__main__":
import os
# Update these paths to match your actual file locations
base_dir = "data/" # Update this
stocks_file = os.path.join(base_dir, "merged/features/stocks_features.parquet")
news_file = os.path.join(base_dir, "marketaux/news/news_latest.parquet")
output_file = os.path.join(base_dir, "merged/features/stocks_features.parquet")
# Check if stocks_features.parquet exists before running
if not os.path.exists(stocks_file):
logger.error(f"Input file missing: {stocks_file}")
print(f"ERROR: Input file missing: {stocks_file}")
exit(1)
# Run the merge with 30-minute tolerance (adjust as needed)
df_result = main(
stocks_file_path=stocks_file,
news_file_path=news_file,
output_file_path=output_file,
time_tolerance_minutes=60*24 # Adjust this based on your needs
) |