app.py

import gradio as gr
import yfinance as yf
import numpy as np
import pandas as pd
from keras.models import load_model
import joblib
import os

# Load the saved components
MODEL_PATH = 'stock_model.keras'
SCALER_PATH = 'scaler.gz'

if os.path.exists(MODEL_PATH) and os.path.exists(SCALER_PATH):
    model = load_model(MODEL_PATH)
    scaler = joblib.load(SCALER_PATH)
else:
    model = None
    scaler = None

def predict_next_day(ticker):
    if model is None or scaler is None:
        return "Error: Model or Scaler files not found."
    
    try:
        # 1. Fetch more data (120d) to accommodate MA50 and RSI calculations
        df = yf.download(ticker, period='120d', interval='1d', multi_level_index=False)
        if df.empty: return f"Error: No data for '{ticker}'."

        # 2. Re-create all 9 features from your notebook
        # Moving Averages
        df['MA7'] = df['Close'].rolling(window=7).mean()
        df['MA21'] = df['Close'].rolling(window=21).mean()
        df['MA50'] = df['Close'].rolling(window=50).mean()
        
        # Volatility & Returns
        df['Volatility'] = df['Close'].rolling(window=7).std()
        df['Daily_Return'] = df['Close'].pct_change()
        
        # Price Range & Volume Trend
        df['Price_Range'] = df['High'] - df['Low']
        df['Volume_MA7'] = df['Volume'].rolling(window=7).mean()
        
        # RSI (Relative Strength Index)
        delta = df['Close'].diff()
        gain = (delta.where(delta > 0, 0)).rolling(window=14).mean()
        loss = (-delta.where(delta < 0, 0)).rolling(window=14).mean()
        rs = gain / loss
        df['RSI'] = 100 - (100 / (1 + rs))
        
        df.dropna(inplace=True)

        # 3. Check if we have enough data left
        if len(df) < 60:
            return "Error: Not enough data for a 60-day prediction window."

        # 4. Prepare features in the EXACT order used in training
        feature_cols = ['Close', 'MA7', 'MA21', 'MA50', 'RSI', 
                        'Volatility', 'Price_Range', 'Daily_Return', 'Volume_MA7']
        
        last_60_days = df[feature_cols].tail(60).values
        
        # 5. Scale and Predict
        scaled_data = scaler.transform(last_60_days)
        # Reshape to (1 sample, 60 time steps, 9 features)
        input_data = np.reshape(scaled_data, (1, 60, 9))
        
        prediction_scaled = model.predict(input_data, verbose=0)

        # 6. Inverse Transform (using 9-column dummy)
        dummy = np.zeros((1, 9))
        dummy[0, 0] = prediction_scaled.flatten()[0]
        prediction_final = scaler.inverse_transform(dummy)[0, 0]

        return f"Predicted Next Closing Price for {ticker}: ₹{prediction_final:.2f}"
    
    except Exception as e:
        return f"Technical Error: {str(e)}"

# Interface
interface = gr.Interface(
    fn=predict_next_day,
    inputs=gr.Textbox(label="Stock Ticker", placeholder="e.g., TCS.NS, RELIANCE.NS, AAPL"),
    outputs=gr.Textbox(label="Forecasted Price"),
    title="📈 StockPulse Predictor",
    description="LSTM model predicting stock price using 9 technical indicators.",
    theme="soft"
)

if __name__ == "__main__":
    interface.launch()