File size: 19,167 Bytes
e8874dd |
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 |
import os
import logging
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, TensorDataset
from torch.optim.lr_scheduler import ReduceLROnPlateau
from sklearn.model_selection import KFold
from sklearn.preprocessing import StandardScaler, LabelEncoder
from sklearn.cluster import DBSCAN
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
from sklearn.base import BaseEstimator, TransformerMixin
import dask.dataframe as dd
import optuna
import numpy as np
import pandas as pd
import pytorch_lightning as pl
import json
# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
# Constants
RANDOM_SEED = 42
N_SPLITS = 5
# Define paths
MODEL_CONFIG_PATH = 'model_config.json'
BEST_MODEL_PATH = 'best_model.pt'
FINAL_MODEL_PATH = 'final_model.pt'
class AdvancedFeatureEngineering(BaseEstimator, TransformerMixin):
"""Custom transformer for advanced feature engineering."""
def __init__(self, numeric_cols):
self.numeric_cols = numeric_cols
def fit(self, X, y=None):
return self
def transform(self, X):
X = X.copy()
# Feature engineering for rank and elo differences with adjusted weighting
X['rank_difference'] = X['winner_rank'] - X['loser_rank']
X['elo_difference'] = (X['winner_eloRating'] - X['loser_eloRating']) * 2 # Further emphasizing ELO ratings
# Optional feature engineering based on available columns
if 'winner_eloRatingDelta' in X.columns and 'loser_eloRatingDelta' in X.columns:
X['elo_differenceDelta'] = (X['winner_eloRatingDelta'] - X['loser_eloRatingDelta']) * 2
X['elo_differenceDelta_abs'] = np.abs(X['elo_differenceDelta'])
# Convert dates to numeric days since the earliest date
try:
X['date'] = pd.to_datetime(X['date'], format='%Y%m%d', errors='coerce')
min_date = X['date'].min()
X['date'] = (X['date'] - min_date).dt.days
except Exception as e:
logging.warning(f"Error converting dates: {str(e)}. Setting 'date' to NaN.")
X['date'] = np.nan
# Convert numeric columns to float, handling errors
for col in self.numeric_cols:
X[col] = pd.to_numeric(X[col], errors='coerce')
return X
def load_data(first_number=2, last_number=15):
"""Load and combine player match data from multiple CSV files."""
dfs = []
core_columns = ['date', 'tournament', 'winner_name', 'winner_rank', 'winner_eloRating',
'loser_name', 'loser_rank', 'loser_eloRating']
optional_columns = ['level', 'bestOf', 'surface', 'indoor', 'speed', 'round',
'winner_seed', 'winner_country_name', 'winner_country_id',
'winner_eloRatingDelta', 'loser_seed', 'loser_country_name',
'loser_country_id', 'loser_eloRatingDelta', 'score', 'outcome', 'loser_entry']
all_columns = core_columns + optional_columns
dtype_dict = {
'loser_entry': 'object',
'outcome': 'object',
'winner_rank': 'float64',
'loser_rank': 'float64',
'winner_eloRating': 'float64',
'loser_eloRating': 'float64',
'bestOf': 'float64',
'speed': 'float64',
'winner_eloRatingDelta': 'float64',
'loser_eloRatingDelta': 'float64',
'indoor': 'float64',
'winner_seed': 'float64',
'loser_seed': 'float64'
}
for i in range(first_number, last_number + 1):
file_path = f'PlayerMatches{i}.csv'
try:
df = dd.read_csv(file_path, low_memory=False, assume_missing=True, dtype=dtype_dict)
missing_core_columns = [col for col in core_columns if col not in df.columns]
if missing_core_columns:
logging.warning(f"Missing core columns in {file_path}: {missing_core_columns}. Skipping this file.")
continue
available_columns = [col for col in all_columns if col in df.columns]
df = df[available_columns].drop_duplicates().compute()
dfs.append(df)
logging.info(f"Loaded {file_path}")
except FileNotFoundError:
logging.warning(f"{file_path} not found. Skipping this file.")
except Exception as e:
logging.warning(f"An error occurred while loading {file_path}: {str(e)}. Skipping this file.")
if not dfs:
raise ValueError("No valid data found.")
combined_df = pd.concat(dfs, ignore_index=True).drop_duplicates()
if combined_df.empty:
raise ValueError("The combined dataframe is empty.")
return combined_df
def determine_column_types(df):
"""Determine numeric and categorical column types in the dataframe."""
numeric_cols = ['winner_rank', 'loser_rank', 'winner_eloRating', 'loser_eloRating']
potential_numeric_cols = ['bestOf', 'speed', 'winner_eloRatingDelta', 'loser_eloRatingDelta', 'indoor', 'winner_seed', 'loser_seed']
for col in potential_numeric_cols:
if col in df.columns:
if pd.api.types.is_numeric_dtype(df[col]) or df[col].str.isnumeric().all():
numeric_cols.append(col)
categorical_cols = [col for col in df.columns if col not in numeric_cols and col != 'date']
return numeric_cols, categorical_cols
def preprocess_data(df, numeric_cols, categorical_cols):
"""Preprocess the dataframe, including encoding categorical variables and handling missing values."""
logging.info(f"Shape before preprocessing: {df.shape}")
label_encoders = {}
for col in categorical_cols:
le = LabelEncoder()
df[col] = df[col].astype(str)
df[col] = le.fit_transform(df[col])
label_encoders[col] = le
feature_engineer = AdvancedFeatureEngineering(numeric_cols)
df = feature_engineer.fit_transform(df)
logging.info(f"Shape after feature engineering: {df.shape}")
# Handle NaN values in numeric and categorical columns
for col in df.columns:
nan_count = df[col].isna().sum()
if nan_count > 0:
logging.warning(f"Column {col} has {nan_count} NaN values")
for col in numeric_cols:
df[col] = pd.to_numeric(df[col], errors='coerce')
df[col] = df[col].fillna(df[col].median())
for col in categorical_cols:
df[col] = df[col].fillna(-1)
# Additional check for NaNs in rank_difference
if df['rank_difference'].isna().any():
logging.error("NaN values found in 'rank_difference' after preprocessing. Identifying rows with NaN values...")
missing_rank_rows = df[df['rank_difference'].isna()]
logging.info(f"Rows with missing 'rank_difference':\n{missing_rank_rows[['winner_rank', 'loser_rank']]}")
df.dropna(subset=['rank_difference'], inplace=True) # Drop rows with NaN in 'rank_difference'
logging.info(f"Shape after dropping NaN rows in 'rank_difference': {df.shape}")
return df, label_encoders
class JointEmbeddedModel(pl.LightningModule):
"""A PyTorch Lightning module for a neural network with categorical embeddings and numeric inputs."""
def __init__(self, categorical_dims, numerical_dim, embedding_dim, hidden_dim, dropout_rate=0.3, learning_rate=1e-3):
super().__init__()
self.embeddings = nn.ModuleList([nn.Embedding(dim, embedding_dim) for dim in categorical_dims])
self.fc1 = nn.Linear(len(categorical_dims) * embedding_dim + numerical_dim, hidden_dim)
self.fc2 = nn.Linear(hidden_dim, hidden_dim // 2)
self.fc3 = nn.Linear(hidden_dim // 2, 1)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(dropout_rate)
self.criterion = nn.MSELoss()
self.learning_rate = learning_rate
def forward(self, x_cat, x_num):
embedded = [emb(x_cat[:, i]) for i, emb in enumerate(self.embeddings)]
x = torch.cat(embedded + [x_num], dim=1)
x = self.dropout(self.relu(self.fc1(x)))
x = self.dropout(self.relu(self.fc2(x)))
return self.fc3(x).squeeze()
def training_step(self, batch, batch_idx):
x_cat, x_num, y = batch
y_hat = self(x_cat, x_num)
loss = self.criterion(y_hat, y)
self.log('train_loss', loss)
return loss
def configure_optimizers(self):
optimizer = optim.AdamW(self.parameters(), lr=self.learning_rate, weight_decay=1e-4) # Using AdamW optimizer
scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=0.5, patience=5)
return {'optimizer': optimizer, 'lr_scheduler': scheduler, 'monitor': 'train_loss'}
def create_dataloader(X, y, batch_size=64):
"""Create a DataLoader for training and evaluation."""
x_cat, x_num = X
x_cat = torch.tensor(x_cat, dtype=torch.long)
x_num = torch.tensor(x_num, dtype=torch.float32)
y = torch.tensor(y, dtype=torch.float32)
dataset = TensorDataset(x_cat, x_num, y)
return DataLoader(dataset, batch_size=batch_size, shuffle=True)
def ensemble_predictions(models, X):
"""Aggregate predictions from an ensemble of models."""
preds = [model.predict(X) for model in models]
return np.mean(preds, axis=0)
def save_model_config(config, path):
"""Save the model configuration to a JSON file."""
with open(path, 'w') as f:
json.dump(config, f)
def load_model_config(path):
"""Load the model configuration from a JSON file."""
with open(path, 'r') as f:
return json.load(f)
def objective(trial):
"""Objective function for hyperparameter optimization with Optuna."""
embedding_dim = trial.suggest_int('embedding_dim', 16, 128)
hidden_dim = trial.suggest_int('hidden_dim', 64, 512)
learning_rate = trial.suggest_float('learning_rate', 1e-6, 1e-2, log=True)
batch_size = trial.suggest_categorical('batch_size', [32, 64, 128, 256])
dropout_rate = trial.suggest_float('dropout_rate', 0.1, 0.5)
model = JointEmbeddedModel(categorical_dims, numerical_dim, embedding_dim, hidden_dim, dropout_rate, learning_rate)
dataloader = create_dataloader(X_train, y_train, batch_size=batch_size)
trainer = pl.Trainer(
max_epochs=20,
accelerator='gpu' if torch.cuda.is_available() else 'cpu', # Use GPU if available
devices=1,
logger=False,
enable_checkpointing=False
)
trainer.fit(model, dataloader)
val_predictions = model(torch.tensor(X_val[0], dtype=torch.long), torch.tensor(X_val[1], dtype=torch.float32)).detach().cpu().numpy()
if np.isnan(y_val).any() or np.isnan(val_predictions).any():
raise ValueError("Validation targets or predictions contain NaN values.")
val_loss = mean_squared_error(y_val, val_predictions)
return val_loss
def analyze_winning_streaks(model, X, df_subset, eps=0.5, min_samples=5, threshold=0.5):
"""Analyze winning streaks using the trained model and clustering techniques."""
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.eval()
x_cat, x_num = X
with torch.no_grad():
embedded = [emb(torch.tensor(x_cat[:, i], dtype=torch.long).to(device)) for i, emb in enumerate(model.embeddings)]
embeddings = torch.cat(embedded, dim=1).cpu().numpy()
outputs = model(torch.tensor(x_cat, dtype=torch.long).to(device),
torch.tensor(x_num, dtype=torch.float32).to(device)).cpu().numpy()
scaler = StandardScaler()
embeddings = scaler.fit_transform(embeddings)
dbscan = DBSCAN(eps=eps, min_samples=min_samples)
labels = dbscan.fit_predict(embeddings)
df_subset['cluster'] = labels
df_subset['predicted_rank_difference'] = outputs
df_subset['easy_draw'] = (df_subset['rank_difference'] - df_subset['predicted_rank_difference']) > threshold
df_subset['hard_draw'] = (df_subset['predicted_rank_difference'] - df_subset['rank_difference']) > threshold
results = df_subset.groupby('winner_name').agg({
'cluster': 'count',
'easy_draw': 'sum',
'hard_draw': 'sum'
}).reset_index()
results['easy_draw_ratio'] = results['easy_draw'] / results['cluster']
results['hard_draw_ratio'] = results['hard_draw'] / results['cluster']
results.sort_values('hard_draw_ratio', ascending=False, inplace=True)
results.to_csv('winning_streak_analysis.csv', index=False)
logging.info(f"Analysis results saved to winning_streak_analysis.csv")
return results
if __name__ == "__main__":
try:
df = load_data()
logging.info(f"Data loaded successfully. Shape: {df.shape}")
numeric_columns, categorical_columns = determine_column_types(df)
logging.info(f"Numeric columns: {numeric_columns}")
logging.info(f"Categorical columns: {categorical_columns}")
df, label_encoders = preprocess_data(df, numeric_columns, categorical_columns)
logging.info(f"Data preprocessed. Shape after preprocessing: {df.shape}")
# Ensure all numeric columns are properly handled
for col in numeric_columns:
if not pd.api.types.is_numeric_dtype(df[col]):
raise ValueError(f"Column {col} contains non-numeric data after preprocessing")
if df.shape[0] < N_SPLITS:
raise ValueError(f"Not enough samples ({df.shape[0]}) for {N_SPLITS}-fold cross-validation.")
X_cat = df[categorical_columns].values
X_num = df[numeric_columns].values.astype(float)
y = df['rank_difference'].values.astype(float)
# Remove NaN values from y
if np.isnan(y).any():
raise ValueError("Target variable contains NaN values.")
logging.info(f"Shape of X_cat: {X_cat.shape}")
logging.info(f"Shape of X_num: {X_num.shape}")
logging.info(f"Shape of y: {y.shape}")
kf = KFold(n_splits=N_SPLITS, shuffle=True, random_state=RANDOM_SEED)
scores = []
for train_index, val_index in kf.split(X_cat):
X_cat_train, X_cat_val = X_cat[train_index], X_cat[val_index]
X_num_train, X_num_val = X_num[train_index], X_num[val_index]
y_train, y_val = y[train_index], y[val_index]
# Additional NaN checks for validation and training sets
if np.isnan(X_cat_train).any() or np.isnan(X_num_train).any() or np.isnan(y_train).any():
raise ValueError("Training data contains NaN values.")
if np.isnan(X_cat_val).any() or np.isnan(X_num_val).any() or np.isnan(y_val).any():
raise ValueError("Validation data contains NaN values.")
X_train = (X_cat_train, X_num_train)
X_val = (X_cat_val, X_num_val)
categorical_dims = [len(label_encoders[col].classes_) for col in categorical_columns]
numerical_dim = len(numeric_columns)
try:
study = optuna.create_study(direction='minimize')
study.optimize(objective, n_trials=100) # Further increased trials for finer parameter search
best_params = study.best_params
logging.info(f"Best Hyperparameters: {best_params}")
# Save the model configuration
model_config = {
'categorical_dims': categorical_dims,
'numerical_dim': numerical_dim,
'embedding_dim': best_params['embedding_dim'],
'hidden_dim': best_params['hidden_dim'],
'dropout_rate': best_params['dropout_rate'],
'learning_rate': best_params['learning_rate']
}
save_model_config(model_config, MODEL_CONFIG_PATH)
model = JointEmbeddedModel(**model_config)
dataloader = create_dataloader(X_train, y_train, batch_size=best_params['batch_size'])
trainer = pl.Trainer(
max_epochs=100, # Further increased max_epochs for deeper training
accelerator='gpu' if torch.cuda.is_available() else 'cpu',
devices=1,
logger=False,
enable_checkpointing=False
)
trainer.fit(model, dataloader)
val_predictions = model(torch.tensor(X_val[0], dtype=torch.long), torch.tensor(X_val[1], dtype=torch.float32)).detach().cpu().numpy()
if np.isnan(val_predictions).any():
raise ValueError("Validation predictions contain NaN values.")
val_loss = mean_squared_error(y_val, val_predictions)
scores.append(val_loss)
# Save the model state
torch.save(model.state_dict(), BEST_MODEL_PATH)
except Exception as e:
logging.error(f"An error occurred during optimization: {str(e)}")
logging.error("Exception details:", exc_info=True)
logging.info(f"Cross-Validation MSE: {np.mean(scores):.4f}")
# Train ensemble models and evaluate
ensemble_models = [
RandomForestRegressor(n_estimators=300, random_state=RANDOM_SEED),
GradientBoostingRegressor(n_estimators=300, random_state=RANDOM_SEED),
LinearRegression()
]
# Check for NaNs in ensemble training data
if np.isnan(np.hstack((X_cat, X_num))).any() or np.isnan(y).any():
raise ValueError("Ensemble training data contains NaN values.")
ensemble_models = [model.fit(np.hstack((X_cat, X_num)), y) for model in ensemble_models]
ensemble_preds = ensemble_predictions(ensemble_models, np.hstack((X_cat, X_num)))
ensemble_mse = mean_squared_error(y, ensemble_preds)
logging.info(f"Ensemble Test MSE: {ensemble_mse:.4f}")
# Load the best model configuration and state for final analysis
if os.path.exists(BEST_MODEL_PATH) and os.path.exists(MODEL_CONFIG_PATH):
model_config = load_model_config(MODEL_CONFIG_PATH)
model = JointEmbeddedModel(**model_config)
model.load_state_dict(torch.load(BEST_MODEL_PATH))
model.eval()
test_predictions = model(torch.tensor(X_cat, dtype=torch.long), torch.tensor(X_num, dtype=torch.float32)).detach().cpu().numpy()
if np.isnan(test_predictions).any():
raise ValueError("Test predictions contain NaN values.")
test_mse = mean_squared_error(y, test_predictions)
logging.info(f"Final Test MSE: {test_mse}")
winning_streak_analysis = analyze_winning_streaks(model, (X_cat, X_num), df)
torch.save(model.state_dict(), FINAL_MODEL_PATH)
logging.info("Script execution completed successfully.")
else:
logging.error("Best model or configuration not found. Ensure training is completed before running analysis.")
except Exception as e:
logging.error(f"An error occurred during script execution: {str(e)}")
logging.error("Exception details:", exc_info=True) |