model_utils.py

import numpy as np
import pandas as pd
import joblib
import onnxruntime as ort

# Load stock data
df_all = pd.read_csv("stocks.csv", index_col=0, parse_dates=True)

def make_prediction(symbol: str, window_size=30):
    if symbol not in df_all.columns:
        raise ValueError(f"{symbol} not found in dataset.")

    series = df_all[symbol].dropna().values
    if len(series) < window_size:
        raise ValueError("Not enough data for prediction.")

    scaler = joblib.load(f"scalers/{symbol}_scaler.pkl")
    scaled_series = scaler.transform(series.reshape(-1, 1)).squeeze()
    input_seq = scaled_series[-window_size:]

    forecast_days = {
        "1_day": 1,
        "1_week": 7,
        "1_month": 30,
        "3_months": 90,
        "6_months": 180
    }

    predictions = {}
    current_input = input_seq.copy()
    session = ort.InferenceSession(f"onnx_models/{symbol}_model.onnx")
    input_name = session.get_inputs()[0].name

    for horizon_name, steps_ahead in forecast_days.items():
        input_window = current_input.copy()
        for _ in range(steps_ahead):
            X = np.array(input_window[-window_size:]).reshape(1, window_size, 1)
            output = session.run(None, {input_name: X})[0]
            next_scaled = output[0][0]
            input_window = np.append(input_window, next_scaled)

        final_scaled = input_window[-1]
        final_pred = scaler.inverse_transform([[final_scaled]])[0][0]
        predictions[horizon_name] = round(float(final_pred), 2)

    return predictions