scripts/merge_embed.py

import os
import sys
import glob
import joblib
import numpy as np
from tqdm import tqdm

def merge_and_stack_folders(folder1_path, folder2_path, output_dir):
    """
    Merge same-name .pkl files from two folders and stack their contents by timestamp.
    """
    # 1. Validate existence of input paths
    if not os.path.isdir(folder1_path):
        print(f"Error: Folder not found -> {folder1_path}")
        return
    if not os.path.isdir(folder2_path):
        print(f"Error: Folder not found -> {folder2_path}")
        return

    # 2. Ensure output directory exists
    os.makedirs(output_dir, exist_ok=True)
    print(f"Input folder 1: {folder1_path}")
    print(f"Input folder 2: {folder2_path}")
    print(f"Output will be saved to: {output_dir}")

    # 3. Find all .pkl files in first folder
    folder1_files = glob.glob(os.path.join(folder1_path, "*.pkl"))
    if not folder1_files:
        print(f"Warning: No .pkl files found in {folder1_path}.")
        return

    print(f"\nFound {len(folder1_files)} files, starting pairwise merging...")
    
    # 4. Traverse files, find pairs and process
    for file1_path in tqdm(folder1_files, desc="Processing file pairs"):
        file_name = os.path.basename(file1_path)
        file2_path = os.path.join(folder2_path, file_name)

        # Check if paired file exists in second folder
        if not os.path.exists(file2_path):
            print(f"\nSkipped: Paired file not found in {folder2_path} for {file_name}")
            continue

        try:
            # Load data from both files
            with open(file1_path, 'rb') as f:
                data1 = joblib.load(f)
            with open(file2_path, 'rb') as f:
                data2 = joblib.load(f)
        except Exception as e:
            print(f"\nError: Failed to load file pair {file_name}: {e}")
            continue

        # Create new dictionary for merged data
        merged_data = {}
        
        # Find common timestamps (keys) in both dictionaries
        common_keys = data1.keys() & data2.keys()

        for key in common_keys:
            vec1 = data1[key]
            vec2 = data2[key]
            
            # Verify vectors meet expectations
            if isinstance(vec1, np.ndarray) and isinstance(vec2, np.ndarray) and \
               vec1.ndim == 2 and vec2.ndim == 2 and \
               vec1.shape[1] == vec2.shape[1]: # Ensure same number of columns
                
                # Vertically stack (N, D) and (M, D) -> (N+M, D)
                # Example: (7, 256) and (7, 256) -> (14, 256)
                stacked_vec = np.vstack((vec1, vec2))
                merged_data[key] = stacked_vec
            else:
                print(f"\nWarning: Data format mismatch at timestamp {key} in file {file_name}, skipped.")

        # Save result if any data was successfully merged
        if merged_data:
            output_path = os.path.join(output_dir, file_name)
            try:
                with open(output_path, 'wb') as f:
                    joblib.dump(merged_data, f)
            except Exception as e:
                print(f"\nError: Failed to save merged file {output_path}: {e}")

def main():
    """
    Main execution function.
    """

    folder1 = "./data/California_ISO/weather/san-francisco"
    folder2 = "./data/California_ISO/weather/san-diego"
    output_folder = "./data/California_ISO/weather/merged_report_embedding"

    merge_and_stack_folders(folder1, folder2, output_folder)
    
    print("\nAll matching files merged and stacked successfully!")
    

if __name__ == "__main__":
    main()