agripredict-analysis / train_catboost.py
adsurkasur's picture
Update AgriPredict Analysis Service
b2c7817 verified
Raw
History Blame Contribute Delete
20.1 kB
"""
CatBoost Model Training Script for AgriPredict
This script trains the CatBoost model using the existing training dataset.
Includes comprehensive accuracy metrics: MAE, RMSE, MAPE, Bias, MASE, R-Squared
"""
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
from catboost import CatBoostRegressor, Pool
from sklearn.model_selection import train_test_split, TimeSeriesSplit
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
import joblib
import os
from typing import Dict, Any, Tuple, Optional
import logging
import json
# Setup logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
class ForecastMetrics:
"""Comprehensive forecast accuracy metrics calculator"""
@staticmethod
def calculate_all_metrics(y_true: np.ndarray, y_pred: np.ndarray,
y_train: Optional[np.ndarray] = None) -> Dict[str, float]:
"""
Calculate all forecast accuracy metrics
Args:
y_true: Actual values
y_pred: Predicted values
y_train: Training data (for MASE calculation)
Returns:
Dictionary with all metrics
"""
y_true = np.array(y_true).flatten()
y_pred = np.array(y_pred).flatten()
# Remove any NaN or infinite values
mask = np.isfinite(y_true) & np.isfinite(y_pred)
y_true = y_true[mask]
y_pred = y_pred[mask]
if len(y_true) == 0:
return {
'mae': np.nan, 'rmse': np.nan, 'mape': np.nan,
'bias': np.nan, 'mase': np.nan, 'r_squared': np.nan
}
# MAE - Mean Absolute Error
mae = mean_absolute_error(y_true, y_pred)
# RMSE - Root Mean Squared Error
rmse = np.sqrt(mean_squared_error(y_true, y_pred))
# MAPE - Mean Absolute Percentage Error (handle zero values)
non_zero_mask = y_true != 0
if np.any(non_zero_mask):
mape = np.mean(np.abs((y_true[non_zero_mask] - y_pred[non_zero_mask]) / y_true[non_zero_mask])) * 100
else:
mape = np.nan
# Bias - Mean Forecast Error (MFE)
bias = np.mean(y_pred - y_true)
# MASE - Mean Absolute Scaled Error
if y_train is not None and len(y_train) > 1:
y_train = np.array(y_train).flatten()
# Naive forecast error (one-step ahead)
naive_errors = np.abs(np.diff(y_train))
scaling_factor = np.mean(naive_errors)
if scaling_factor > 0:
mase = mae / scaling_factor
else:
mase = np.nan
else:
mase = np.nan
# R-Squared
r_squared = r2_score(y_true, y_pred)
return {
'mae': round(float(mae), 4),
'rmse': round(float(rmse), 4),
'mape': round(float(mape), 4) if not np.isnan(mape) else None,
'bias': round(float(bias), 4),
'mase': round(float(mase), 4) if not np.isnan(mase) else None,
'r_squared': round(float(r_squared), 4)
}
class CatBoostTrainer:
"""CatBoost model trainer for agricultural demand forecasting"""
def __init__(self):
self.model = None
self.feature_names = None
self.metrics = None
self.training_config = None
def load_training_data(self, filepath: str = None) -> pd.DataFrame:
"""
Load training data from CSV file
Args:
filepath: Path to training data CSV
Returns:
DataFrame with training data
"""
if filepath is None:
# Default to the existing training data
filepath = os.path.join(os.path.dirname(__file__), '..', 'data', 'catboost_training_data.csv')
if not os.path.exists(filepath):
filepath = os.path.join(os.path.dirname(__file__), '..', '..', 'data', 'catboost_training_data.csv')
if not os.path.exists(filepath):
logger.warning(f"Training data not found at {filepath}, generating synthetic data")
return self.generate_artificial_data(n_samples=5000)
logger.info(f"Loading training data from {filepath}")
df = pd.read_csv(filepath)
# Parse date column
if 'date' in df.columns:
df['date'] = pd.to_datetime(df['date'])
logger.info(f"Loaded {len(df)} samples with {len(df.columns)} features")
return df
def generate_artificial_data(self, n_samples: int = 5000) -> pd.DataFrame:
"""
Generate artificial agricultural data for training
Args:
n_samples: Number of samples to generate
Returns:
DataFrame with artificial agricultural data
"""
logger.info(f"Generating {n_samples} artificial data samples")
# Generate date range
start_date = datetime(2023, 1, 1)
dates = [start_date + timedelta(days=i) for i in range(n_samples)]
np.random.seed(42) # For reproducible results
data = []
for date in dates:
# Seasonal patterns
day_of_year = date.timetuple().tm_yday
seasonal_factor = 1 + 0.3 * np.sin(2 * np.pi * day_of_year / 365)
# Base demand with seasonal variation
base_quantity = np.random.normal(100, 20) * seasonal_factor
# Price influenced by season and demand
base_price = 25 + 5 * np.sin(2 * np.pi * day_of_year / 365)
price_noise = np.random.normal(0, 2)
price = base_price + price_noise
# Add some correlation between price and quantity
quantity_noise = np.random.normal(0, 15)
quantity = base_quantity + quantity_noise - 0.1 * (price - 25)
# Ensure positive values
quantity = max(1, quantity)
price = max(5, price)
data.append({
'date': date,
'quantity': round(quantity, 2),
'price': round(price, 2),
'day_of_week': date.weekday(),
'month': date.month,
'day_of_month': date.day,
'quarter': (date.month - 1) // 3 + 1,
'is_weekend': 1 if date.weekday() >= 5 else 0,
'season': self._get_season(date.month)
})
df = pd.DataFrame(data)
# Add lag features
for lag in [1, 7, 14, 30]:
df[f'price_lag_{lag}'] = df['price'].shift(lag)
df[f'quantity_lag_{lag}'] = df['quantity'].shift(lag)
# Add rolling statistics
for window in [7, 14, 30]:
df[f'price_rolling_mean_{window}'] = df['price'].rolling(window).mean()
df[f'price_rolling_std_{window}'] = df['price'].rolling(window).std()
df[f'quantity_rolling_mean_{window}'] = df['quantity'].rolling(window).mean()
# Add price change features
df['price_change'] = df['price'].pct_change()
df['price_change_7d'] = df['price'].pct_change(7)
# Drop rows with NaN values
df = df.dropna().reset_index(drop=True)
logger.info(f"Generated dataset with {len(df)} samples and {len(df.columns)} features")
return df
def _get_season(self, month: int) -> str:
"""Get season based on month"""
if month in [12, 1, 2]:
return 'winter'
elif month in [3, 4, 5]:
return 'spring'
elif month in [6, 7, 8]:
return 'summer'
else:
return 'fall'
def prepare_features(self, df: pd.DataFrame, target_col: str = 'target_quantity') -> tuple:
"""
Prepare features for training
Args:
df: Input DataFrame
target_col: Name of target column
Returns:
Tuple of (X, y, feature_names, categorical_features)
"""
# Define columns to exclude from features
exclude_cols = ['date', 'target_quantity', 'product_name', 'region_name', 'product_category']
# Check if target column exists
if target_col not in df.columns:
# Fallback to price if target_quantity doesn't exist
target_col = 'price' if 'price' in df.columns else df.columns[-1]
logger.warning(f"target_quantity not found, using {target_col} as target")
# Define feature columns
feature_cols = [col for col in df.columns if col not in exclude_cols and col != target_col]
# Identify categorical features
categorical_features = []
for col in feature_cols:
if df[col].dtype == 'object' or df[col].dtype.name == 'category':
categorical_features.append(col)
# Remove categorical features for now (CatBoost handles them, but we'll use numeric only)
feature_cols = [col for col in feature_cols if col not in categorical_features]
# Prepare features and target
X = df[feature_cols].copy()
y = df[target_col].copy()
# Handle any remaining NaN values
X = X.fillna(X.median())
logger.info(f"Prepared {len(feature_cols)} features for training, target: {target_col}")
return X, y, feature_cols, categorical_features
def train_model(self, X_train, y_train, X_val=None, y_val=None, **kwargs) -> CatBoostRegressor:
"""
Train CatBoost model
Args:
X_train: Training features
y_train: Training target
X_val: Validation features (optional)
y_val: Validation target (optional)
**kwargs: Additional CatBoost parameters
Returns:
Trained CatBoost model
"""
# Default parameters optimized for demand forecasting
default_params = {
'iterations': 1000,
'learning_rate': 0.05,
'depth': 8,
'loss_function': 'MAE',
'eval_metric': 'MAE',
'random_seed': 42,
'verbose': 100,
'early_stopping_rounds': 100,
'l2_leaf_reg': 3,
'border_count': 128,
'thread_count': -1
}
# Update with custom parameters
default_params.update(kwargs)
self.training_config = default_params.copy()
# Create model
model = CatBoostRegressor(**default_params)
# Prepare data
train_pool = Pool(X_train, y_train)
if X_val is not None and y_val is not None:
val_pool = Pool(X_val, y_val)
model.fit(train_pool, eval_set=val_pool, use_best_model=True)
else:
model.fit(train_pool)
self.model = model
self.feature_names = list(X_train.columns)
logger.info(f"Trained CatBoost model with {model.tree_count_} trees")
return model
def evaluate_model(self, X_test, y_test, y_train=None) -> Dict[str, float]:
"""
Evaluate model performance with comprehensive metrics
Args:
X_test: Test features
y_test: Test target
y_train: Training target (for MASE calculation)
Returns:
Dictionary with evaluation metrics
"""
if self.model is None:
raise ValueError("Model not trained yet")
# Make predictions
y_pred = self.model.predict(X_test)
# Calculate all metrics
self.metrics = ForecastMetrics.calculate_all_metrics(
y_true=y_test,
y_pred=y_pred,
y_train=y_train
)
logger.info(f"Model Evaluation Metrics:")
logger.info(f" MAE: {self.metrics['mae']}")
logger.info(f" RMSE: {self.metrics['rmse']}")
logger.info(f" MAPE: {self.metrics['mape']}%")
logger.info(f" Bias: {self.metrics['bias']}")
logger.info(f" MASE: {self.metrics['mase']}")
logger.info(f" R²: {self.metrics['r_squared']}")
return self.metrics
def save_model(self, filepath: str):
"""
Save trained model to file with metadata
Args:
filepath: Path to save the model
"""
if self.model is None:
raise ValueError("Model not trained yet")
# Create directory if it doesn't exist
os.makedirs(os.path.dirname(filepath) if os.path.dirname(filepath) else '.', exist_ok=True)
# Prepare model data with metadata
model_data = {
'model': self.model,
'feature_names': self.feature_names,
'metrics': self.metrics,
'training_config': self.training_config,
'training_date': datetime.now().isoformat(),
'version': '2.0'
}
# Save model using joblib
joblib.dump(model_data, filepath)
# Also save the CatBoost native format for faster loading
native_path = filepath.replace('.pkl', '.cbm')
self.model.save_model(native_path)
# Save metrics to JSON for easy access
metrics_path = filepath.replace('.pkl', '_metrics.json')
with open(metrics_path, 'w') as f:
json.dump({
'metrics': self.metrics,
'feature_names': self.feature_names,
'training_date': model_data['training_date'],
'training_config': {k: str(v) for k, v in (self.training_config or {}).items()}
}, f, indent=2)
logger.info(f"Model saved to {filepath}")
logger.info(f"Native model saved to {native_path}")
logger.info(f"Metrics saved to {metrics_path}")
def load_model(self, filepath: str):
"""
Load trained model from file
Args:
filepath: Path to the saved model
"""
if not os.path.exists(filepath):
raise FileNotFoundError(f"Model file not found: {filepath}")
# Load model
model_data = joblib.load(filepath)
self.model = model_data['model']
self.feature_names = model_data.get('feature_names', [])
self.metrics = model_data.get('metrics', {})
self.training_config = model_data.get('training_config', {})
logger.info(f"Model loaded from {filepath}")
logger.info(f"Model metrics: {self.metrics}")
def predict(self, features: pd.DataFrame) -> np.ndarray:
"""
Make predictions with trained model
Args:
features: Input features
Returns:
Predictions array
"""
if self.model is None:
raise ValueError("Model not trained or loaded yet")
# Ensure features are in correct order
if self.feature_names:
# Only use features that exist in both
available_features = [f for f in self.feature_names if f in features.columns]
missing_features = [f for f in self.feature_names if f not in features.columns]
if missing_features:
logger.warning(f"Missing features: {missing_features[:5]}...")
# Add missing features with default values
for f in missing_features:
features[f] = 0
features = features[self.feature_names]
return self.model.predict(features)
def get_feature_importance(self, top_n: int = 20) -> Dict[str, float]:
"""
Get feature importance from trained model
Args:
top_n: Number of top features to return
Returns:
Dictionary of feature names and their importance scores
"""
if self.model is None:
raise ValueError("Model not trained yet")
importance = self.model.get_feature_importance()
feature_importance = dict(zip(self.feature_names, importance))
# Sort by importance
sorted_importance = dict(sorted(feature_importance.items(), key=lambda x: x[1], reverse=True)[:top_n])
return sorted_importance
def main():
"""Main training function"""
logger.info("=" * 60)
logger.info("Starting CatBoost Model Training for AgriPredict")
logger.info("=" * 60)
# Initialize trainer
trainer = CatBoostTrainer()
# Load training data (use existing CSV or generate synthetic)
df = trainer.load_training_data()
logger.info(f"Dataset shape: {df.shape}")
logger.info(f"Columns: {list(df.columns)[:10]}...")
# Prepare features
X, y, feature_names, categorical_features = trainer.prepare_features(df)
logger.info(f"Feature matrix shape: {X.shape}")
logger.info(f"Target shape: {y.shape}")
logger.info(f"Target statistics: mean={y.mean():.2f}, std={y.std():.2f}, min={y.min():.2f}, max={y.max():.2f}")
# Split data (use time-based split for time series data)
# Reserve last 20% for testing
split_idx = int(len(X) * 0.8)
X_train_full, X_test = X.iloc[:split_idx], X.iloc[split_idx:]
y_train_full, y_test = y.iloc[:split_idx], y.iloc[split_idx:]
# Further split training data for validation
val_split_idx = int(len(X_train_full) * 0.8)
X_train, X_val = X_train_full.iloc[:val_split_idx], X_train_full.iloc[val_split_idx:]
y_train, y_val = y_train_full.iloc[:val_split_idx], y_train_full.iloc[val_split_idx:]
logger.info(f"Train set: {len(X_train)} samples")
logger.info(f"Validation set: {len(X_val)} samples")
logger.info(f"Test set: {len(X_test)} samples")
# Train model
logger.info("-" * 40)
logger.info("Training CatBoost model...")
model = trainer.train_model(X_train, y_train, X_val, y_val)
# Evaluate model on test set
logger.info("-" * 40)
logger.info("Evaluating model on test set...")
metrics = trainer.evaluate_model(X_test, y_test, y_train=y_train)
# Get feature importance
logger.info("-" * 40)
logger.info("Top 10 Feature Importance:")
importance = trainer.get_feature_importance(top_n=10)
for feat, imp in importance.items():
logger.info(f" {feat}: {imp:.2f}")
# Save model
model_dir = os.path.join(os.path.dirname(__file__), 'models')
os.makedirs(model_dir, exist_ok=True)
model_path = os.path.join(model_dir, "catboost_model.pkl")
trainer.save_model(model_path)
logger.info("=" * 60)
logger.info("Training completed successfully!")
logger.info(f"Model saved to: {model_path}")
logger.info("=" * 60)
# Print final summary
print("\n" + "=" * 60)
print("TRAINING SUMMARY")
print("=" * 60)
print(f"Total samples: {len(df)}")
print(f"Features used: {len(feature_names)}")
print(f"Trees in model: {model.tree_count_}")
print("\nTest Set Metrics:")
print(f" MAE: {metrics['mae']:.4f}")
print(f" RMSE: {metrics['rmse']:.4f}")
print(f" MAPE: {metrics['mape']:.2f}%" if metrics['mape'] else " MAPE: N/A")
print(f" Bias: {metrics['bias']:.4f}")
print(f" MASE: {metrics['mase']:.4f}" if metrics['mase'] else " MASE: N/A")
print(f" R²: {metrics['r_squared']:.4f}")
print("=" * 60)
return trainer
if __name__ == "__main__":
trained_trainer = main()