target_train.py

#!/usr/bin/env python3
"""
AgGPT-20 Target Trainer
Simple single-file training script that creates individual .feather models

Usage:
    python target_train.py [file_path]
    
If no file path is provided, it will use extra_corpora/example.txt by default.
"""

import os
import glob
from train import AgGPTTrainer, MiniModelTrainer
from feather import FeatherManager

def get_next_target_model_number(models_dir="models"):
    """Find the next available target_AgGPT_X.feather number"""
    if not os.path.exists(models_dir):
        os.makedirs(models_dir)
    
    # Find all existing target_AgGPT_*.feather files
    pattern = os.path.join(models_dir, "target_AgGPT_*.feather")
    existing_files = glob.glob(pattern)
    
    if not existing_files:
        return 1
    
    # Extract numbers from existing files
    numbers = []
    for file_path in existing_files:
        filename = os.path.basename(file_path)
        # Extract number from target_AgGPT_X.feather
        try:
            number_str = filename.split('_')[2].split('.')[0]
            numbers.append(int(number_str))
        except (IndexError, ValueError):
            continue
    
    return max(numbers) + 1 if numbers else 1

def train_target_model(file_path="extra_corpora/example.txt", models_dir="models"):
    """Train a single target model from a text file"""
    print("AgGPT-20 Target Trainer")
    print("=" * 40)
    
    # Check if file exists
    if not os.path.exists(file_path):
        print(f"Error: Training file '{file_path}' not found!")
        print(f"Please create the file or specify a different path.")
        return False
    
    # Check file size
    file_size = os.path.getsize(file_path)
    if file_size == 0:
        print(f"Error: Training file '{file_path}' is empty!")
        return False
    
    print(f"Training file: {file_path}")
    print(f"File size: {file_size} bytes")
    
    # Create models directory if it doesn't exist
    if not os.path.exists(models_dir):
        os.makedirs(models_dir)
    
    # Initialize trainer components
    feather_manager = FeatherManager(models_dir)
    trainer = AgGPTTrainer(models_dir)
    mini_trainer = MiniModelTrainer(feather_manager)
    
    # Load training data
    print(f"Loading training data from {file_path}...")
    try:
        training_pairs = trainer.load_training_data(file_path)
    except Exception as e:
        print(f"Error loading training data: {e}")
        return False
    
    if not training_pairs:
        print("No valid training pairs found in the file!")
        print("Please check the file format:")
        print("user: [your message]")
        print("<pad>")
        print("ai: [response]")
        print("<eos>")
        return False
    
    print(f"Found {len(training_pairs)} training pairs")
    
    # Train the mini-model
    print("Training mini-model...")
    try:
        mini_model = mini_trainer.train_mini_model(training_pairs)
    except Exception as e:
        print(f"Error during training: {e}")
        return False
    
    if not mini_model:
        print("Failed to create mini-model!")
        return False
    
    # Get next available model number
    model_number = get_next_target_model_number(models_dir)
    model_filename = f"target_AgGPT_{model_number}.feather"
    
    # Save the model with custom filename
    print(f"Saving model as {model_filename}...")
    try:
        feather_manager.save_mini_model(mini_model, model_number, model_filename)
        model_path = os.path.join(models_dir, model_filename)
        print(f"✅ Successfully created: {model_path}")
        
    except Exception as e:
        print(f"Error saving model: {e}")
        return False
    
    # Display model statistics
    print("\nModel Statistics:")
    print(f"  Training samples: {mini_model.get('training_samples', 0)}")
    print(f"  Patterns: {len(mini_model.get('patterns', []))}")
    print(f"  Responses: {len(mini_model.get('responses', []))}")
    print(f"  Confidence: {mini_model.get('confidence', 0.0):.3f}")
    print(f"  Keywords: {len(mini_model.get('keywords', []))}")
    
    print(f"\n🎉 Target model training complete!")
    print(f"The model will be automatically discovered by chat.py")
    print("Run 'python chat.py' to test your new model!")
    
    return True

def main():
    import sys
    
    # Check command line arguments
    if len(sys.argv) > 1:
        file_path = sys.argv[1]
    else:
        file_path = "extra_corpora/example.txt"
    
    success = train_target_model(file_path)
    
    if not success:
        print("\n❌ Training failed!")
        sys.exit(1)

if __name__ == "__main__":
    main()