scripts/embed_static.py

import json
import joblib
import numpy as np
import torch
import os
from transformers import AutoModel

# Set proxy (replace with your proxy address and port)
os.environ['HTTP_PROXY'] = 'http://localhost:1080'
os.environ['HTTPS_PROXY'] = 'http://localhost:1080'

def create_static_embeddings(
    input_path="./data/Bear_room/static_info.json",
    output_path="./data/Bear_room/static_info_embeddings.pkl",
):
    """
    Loads static information from a JSON file, generates embeddings for the text fields,
    and saves the result as a pickle file.

    Args:
        input_path (str): Path to the input static_info.json file.
        output_path (str): Path to save the output .pkl file with embeddings.
    """

    # --- 2. Initialize Model ---
    print("Initializing embedding model...")
    # Set up device (use GPU if available)
    device = "cuda:0" if torch.cuda.is_available() else "cpu"
    print(f"Using device: {device}")
    
    # Load the pre-trained model
    model = AutoModel.from_pretrained(
        "jinaai/jina-embeddings-v3", 
        trust_remote_code=True
    ).to(device=device)

    # --- 3. Load and Prepare Data ---
    print(f"Loading data from '{input_path}'...")
    with open(input_path, "r") as f:
        static_info = json.load(f)

    # Start with general texts
    texts_to_embed = [
        static_info['general_info'], 
        static_info['downtime_prompt']
    ]
    
    # Create a structure to remember the path to each channel text
    # e.g., [('CFH_HQ', 'location_description'), ('CFH_HQ', 'panel_type'), ...]
    channel_info_paths = []
    for channel_name, details_dict in static_info['channel_info'].items():
        for sub_key, text_value in details_dict.items():
            texts_to_embed.append(text_value)
            channel_info_paths.append((channel_name, sub_key))

    print(f"Found {len(texts_to_embed)} text snippets to embed.")

    # --- 4. Generate Embeddings ---
    print("Generating embeddings...")
    embeddings = model.encode(
        texts_to_embed,
        truncate_dim=256  # Truncate to 256 dimensions as in the notebook
    )
    print(f"Embeddings generated with shape: {embeddings.shape}")

    # --- 5. Replace Text with Embeddings in the Dictionary ---
    print("Replacing text data with embeddings...")
    # The original static_info dictionary is modified in place
    static_info['general_info'] = embeddings[0:1, :]
    static_info['downtime_prompt'] = embeddings[1:2, :]

    # Start from the 2nd index for channel info embeddings
    channel_embeddings_start_index = 2
    channel_embeddings_dict = {key: [] for key in static_info['channel_info'].keys()}
    for i, (channel_name, sub_key) in enumerate(channel_info_paths):
        # Calculate the correct index in the flat embeddings array
        embedding_index = channel_embeddings_start_index + i
        # The slice keeps the result as a 2D array (1, 256)
        channel_embeddings_dict[channel_name].append(embeddings[embedding_index:embedding_index+1, :])

    # print(channel_embeddings_dict['weather_large'][0].shape)  # Example to check the shape of the first embedding

    for channel_name, embeddings_list in channel_embeddings_dict.items():
        # Stack the embeddings for each channel into a single embedding
        stacked_embedding = np.squeeze(np.stack(embeddings_list, axis=0))
        # Replace the list of embeddings with the stacked embedding
        static_info['channel_info'][channel_name] = stacked_embedding

        print(f"Channel '{channel_name}' embeddings shape: {stacked_embedding.shape}")

    # --- 6. Save the Result ---
    # Ensure the output directory exists
    output_dir = os.path.dirname(output_path)
    if output_dir:
        os.makedirs(output_dir, exist_ok=True)
        
    print(f"Saving embeddings to '{output_path}'...")
    with open(output_path, "wb") as f:
        joblib.dump(static_info, f)

    print("Process completed successfully!")


if __name__ == "__main__":
    create_static_embeddings()