app.py

from flask import Flask, request, jsonify
import pandas as pd
import numpy as np
import baostock as bs
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_error
from neuralprophet import NeuralProphet, set_log_level
import torch
import torch.nn as nn
from torch.optim import Adam
import os

# Initialize Flask app
app = Flask(__name__)

# Set log level to suppress unnecessary warnings
set_log_level("ERROR")

# Baostock API login
lg = bs.login()
if lg.error_code != '0':
    raise ConnectionError(f"Baostock login failed. Error code: {lg.error_code}, Error message: {lg.error_msg}")

# Collect historical data
def get_historical_data(start_date, end_date):
    data = bs.query_history_k_data_plus(
        "sz.000001",  # Shanghai Composite Index
        "date,open,high,low,close,volume",
        start_date=start_date,
        end_date=end_date,
        frequency="d"
    )
    if data.error_code != '0':
        raise ValueError(f"Error in fetching data: {data.error_msg}")

    data_list = []
    while data.next():
        data_list.append(data.get_row_data())
    data_df = pd.DataFrame(data_list, columns=data.fields)

    # Convert relevant columns to numeric type
    data_df[['open', 'close', 'high', 'low', 'volume']] = data_df[['open', 'close', 'high', 'low', 'volume']].apply(pd.to_numeric, errors='coerce')
    return data_df.dropna()

# Filter stocks based on conditions
def filter_stocks(data_df):
    data_df = data_df[(data_df["open"] >= 0.98 * data_df["close"].shift(1).fillna(0)) & (data_df["open"] <= 1.02 * data_df["close"].shift(1).fillna(0))]
    data_df = data_df[(data_df["high"] == data_df["close"]) & (data_df["low"] == data_df["close"]) & (data_df["open"] != 0) & (data_df["close"] != 0)]
    return data_df

# Prepare the training and validation data
data_df = get_historical_data("2005-05-30", "2024-01-31")
filtered_df = filter_stocks(data_df)

if filtered_df.empty:
    raise ValueError("Filtered dataset is empty. Please adjust the filtering conditions.")

train_data, val_data = train_test_split(filtered_df, test_size=0.2, random_state=42)

# Define custom model
class CustomModel(nn.Module):
    def __init__(self):
        super(CustomModel, self).__init__()
        self.neural_prophet = NeuralProphet(
            n_forecasts=1,
            n_lags=30,
            n_changepoints=10,
            changepoints_range=0.8,
            learning_rate=1e-3,
            optimizer=Adam,
        )

    def predict(self, df):
        future = self.neural_prophet.make_future_dataframe(df, periods=1)
        forecast = self.neural_prophet.predict(future)
        return forecast['yhat1'].values

# Instantiate model
model = CustomModel()

# Prepare data for prediction
def prepare_data(date):
    data_df = get_historical_data("2005-05-30", date)
    filtered_df = filter_stocks(data_df)
    if filtered_df.empty:
        return pd.DataFrame()  # Return empty DataFrame if no data matches the filter

    # Scale the data using MinMaxScaler
    scaler = MinMaxScaler()
    filtered_df[['open', 'high', 'low', 'close', 'volume']] = scaler.fit_transform(filtered_df[['open', 'high', 'low', 'close', 'volume']])
    return filtered_df

# Define a route to predict the top 5 stock codes
@app.route('/predict', methods=['POST'])
def predict():
    try:
        date = request.json['date']
        data_df = prepare_data(date)
        if data_df.empty:
            return jsonify({'error': 'No data available for the given date'}), 400

        y_pred = model.predict(data_df)
        top_5_stocks = y_pred[:5]  # Assuming y_pred contains the predicted values for stocks
        return jsonify({'top_5_stocks': top_5_stocks.tolist()})
    except Exception as e:
        return jsonify({'error': str(e)}), 500

# Run the Flask app
if __name__ == '__main__':
    app.run(debug=True)

# Logout from Baostock API
bs.logout()