app.py

import os
import gradio as gr
import tensorflow as tf
import joblib
import numpy as np
import pandas as pd
import yfinance as yf
from huggingface_hub import hf_hub_download

# --- 0. Force CPU-only mode for TensorFlow ---
# This prevents TensorFlow from trying to allocate GPU memory on a CPU-only instance.
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'

# --- 1. Define Constants and Download Model/Scalers from Hugging Face Hub ---
MODEL_REPO = "munem420/stock-forecaster-lstm"
MODEL_FILENAME = "model_lstm.keras"
SCALER_FILENAME = "scalers.joblib"

print("--- Downloading model and scalers from Hugging Face Hub ---")
try:
    model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILENAME)
    scalers_path = hf_hub_download(repo_id=MODEL_REPO, filename=SCALER_FILENAME)
    print("✅ Files downloaded successfully.")
except Exception as e:
    print(f"❌ Critical Error: Could not download files from the Hub. {e}")
    # Set paths to None so the app knows that loading failed.
    model_path, scalers_path = None, None

# --- 2. Load the Model and Scalers into Memory ---
loaded_model_lstm = None
loaded_scalers = None

if model_path and os.path.exists(model_path):
    try:
        loaded_model_lstm = tf.keras.models.load_model(model_path)
        print("✅ TensorFlow model loaded successfully.")
    except Exception as e:
        print(f"❌ Critical Error: Could not load the TensorFlow model. {e}")

if scalers_path and os.path.exists(scalers_path):
    try:
        loaded_scalers = joblib.load(scalers_path)
        print("✅ Scalers loaded successfully.")
    except Exception as e:
        print(f"❌ Critical Error: Could not load the scalers file. {e}")

# --- 3. The Core Forecasting Function ---
def forecast_stock(input_name: str, input_width: int = 60) -> str:
    """
    Fetches live stock data, preprocesses it, and returns a prediction string.
    """
    # Fail fast if the model/scalers didn't load during startup
    if not loaded_model_lstm or not loaded_scalers:
        return "Error: Model or scalers are not loaded. The backend may have failed to start correctly. Check the Space logs."

    ticker = input_name.strip().upper()
    if not ticker:
        return "Error: Please enter a stock ticker."

    print(f"\n--- Generating forecast for {ticker} ---")
    
    # Fetch recent data using yfinance
    try:
        # Fetch more than needed to ensure we have enough valid trading days
        data_df = yf.download(ticker, period="200d", progress=False)
        if data_df.empty:
             return f"Error: No data found for ticker '{ticker}'. It may be delisted or an invalid symbol."
    except Exception as e:
        return f"Error fetching data for '{ticker}': {e}"

    if len(data_df) < input_width:
        return f"Error: Not enough historical data for {ticker}. Need {input_width} days, but only found {len(data_df)}."

    # Prepare the data for the model
    recent_data = data_df.tail(input_width)
    close_prices = recent_data['Close'].values.reshape(-1, 1)

    # Find the correct scaler. The original model was trained on specific stocks.
    # We try to find a matching scaler, otherwise, we use a default as a fallback.
    scaler = loaded_scalers.get(ticker)
    if not scaler:
        print(f"Warning: No specific scaler found for {ticker}. Using ZURVY's scaler as a fallback.")
        scaler = loaded_scalers.get('ZURVY')
        if not scaler:
            return "Error: Critical failure. The default 'ZURVY' scaler could not be found."

    # Scale the data and make a prediction
    try:
        scaled_data = scaler.transform(close_prices)
        X_pred = scaled_data.reshape(1, input_width, 1) # Reshape for LSTM: [batch, timesteps, features]

        prediction_scaled = loaded_model_lstm.predict(X_pred, verbose=0)[0][0]
        prediction_actual = scaler.inverse_transform(np.array([[prediction_scaled]]))[0][0]
    except Exception as e:
        return f"An error occurred during model prediction: {e}"

    # Format the final result
    last_close = recent_data['Close'].iloc[-1]
    result_str = (
        f"Forecast for: {ticker}\n"
        f"Last Close Price: ${last_close:.2f}\n"
        f"Predicted Next Day's Close: ${prediction_actual:.2f}"
    )
    print(result_str)
    return result_str

# --- 4. Create the Gradio Interface and API Endpoint ---
def predict_api(ticker_symbol: str) -> str:
    """A simple wrapper for the main forecast function to be exposed as an API."""
    return forecast_stock(ticker_symbol)

with gr.Blocks(title="Stock Forecaster Backend") as app:
    gr.Markdown("## Stock Forecaster Backend\nThis Gradio app serves the API for the React frontend.")
    
    # These components are not visible but are required to create the API endpoint
    ticker_input = gr.Textbox(label="Stock Ticker", visible=False)
    output_text = gr.Textbox(label="Forecast", visible=False)
    
    # This creates the API endpoint at /run/predict
    ticker_input.submit(
        fn=predict_api, 
        inputs=[ticker_input], 
        outputs=[output_text], 
        api_name="predict"
    )

# --- 5. Mount the static React build directory to be served ---
# This requires a recent version of Gradio (e.g., 4.x), specified in README.md
app = gr.mount_static_directory(app, "build")

# --- 6. Launch the Gradio App ---
if __name__ == "__main__":
    app.launch()