stockstream / app.py
mistermprah's picture
Update app.py
f6a2537 verified
raw
history blame
6.45 kB
import streamlit as st
import numpy as np
import pandas as pd
import yfinance as yf
from datetime import datetime
from tensorflow.keras.models import load_model
from joblib import load
# Load the saved LSTM model and scaler
lstm_model = load_model('lstm_model.h5')
scaler = load('scaler.joblib')
# Dictionary of stock tickers and their full names
stock_dict = {
'GOOG': 'Alphabet Inc.',
'AAPL': 'Apple Inc.',
'TSLA': 'Tesla, Inc.',
'AMZN': 'Amazon.com, Inc.',
'MSFT': 'Microsoft Corporation'
}
# Function to get the last row of stock data
def get_last_stock_data(ticker):
try:
start_date = '2010-01-01'
end_date = datetime.now().strftime('%Y-%m-%d')
data = yf.download(ticker, start=start_date, end=end_date)
last_row = data.iloc[-1]
return last_row.to_dict()
except Exception as e:
return str(e)
# Function to make predictions
def predict_stock_price(ticker, open_price, close_price):
try:
start_date = '2010-01-01'
end_date = datetime.now().strftime('%Y-%m-%d')
data = yf.download(ticker, start=start_date, end=end_date)
# Prepare the data
data = data[['Close']]
dataset = data.values
scaled_data = scaler.transform(dataset)
# Append the user inputs as the last row in the data
user_input = np.array([[close_price]])
user_input_scaled = scaler.transform(user_input)
scaled_data = np.vstack([scaled_data, user_input_scaled])
# Prepare the data for LSTM
x_test_lstm = []
for i in range(60, len(scaled_data)):
x_test_lstm.append(scaled_data[i-60:i])
x_test_lstm = np.array(x_test_lstm)
x_test_lstm = np.reshape(x_test_lstm, (x_test_lstm.shape[0], x_test_lstm.shape[1], 1))
# LSTM Predictions
lstm_predictions = lstm_model.predict(x_test_lstm)
lstm_predictions = scaler.inverse_transform(lstm_predictions)
next_day_lstm_price = lstm_predictions[-1][0]
result = f"Predicted future price for {ticker}: ${next_day_lstm_price:.2f}"
return result
except Exception as e:
return str(e)
# Function to predict next month's price
def predict_next_month_price(ticker, close_price):
try:
start_date = '2010-01-01'
end_date = datetime.now().strftime('%Y-%m-%d')
data = yf.download(ticker, start=start_date, end=end_date)
# Prepare the data
data = data[['Close']]
dataset = data.values
scaled_data = scaler.transform(dataset)
# Append the user inputs as the last row in the data
user_input = np.array([[close_price]])
user_input_scaled = scaler.transform(user_input)
scaled_data = np.vstack([scaled_data, user_input_scaled])
# Prepare the data for LSTM
x_test_lstm = []
for i in range(60, len(scaled_data)):
x_test_lstm.append(scaled_data[i-60:i])
x_test_lstm = np.array(x_test_lstm)
x_test_lstm = np.reshape(x_test_lstm, (x_test_lstm.shape[0], x_test_lstm.shape[1], 1))
# Predicting the next 30 days
predictions = []
for _ in range(30):
pred = lstm_model.predict(x_test_lstm[-1].reshape(1, 60, 1))
predictions.append(pred[0])
new_input = np.append(x_test_lstm[-1][1:], pred)
x_test_lstm = np.append(x_test_lstm, new_input.reshape(1, 60, 1), axis=0)
predictions = np.array(predictions)
next_month_predictions = scaler.inverse_transform(predictions)
next_month_price = next_month_predictions[-1][0]
result = f"Predicted price for {ticker} next month: ${next_month_price:.2f}"
return result
except Exception as e:
return str(e)
# Function to display historical data
def display_historical_data(ticker):
try:
start_date = '2010-01-01'
end_date = datetime.now().strftime('%Y-%m-%d')
data = yf.download(ticker, start=start_date, end=end_date)
return data.tail(30).iloc[::-1] # Reverse to have the latest date on top
except Exception as e:
return str(e)
# Streamlit interface
st.title("Stockstream")
# Sidebar for adding new stocks
st.sidebar.header("Add a New Stock Ticker")
new_ticker = st.sidebar.text_input("Stock Ticker", value="")
new_full_name = st.sidebar.text_input("Full Name", value="")
if st.sidebar.button("Add Stock Ticker"):
if new_ticker and new_full_name:
stock_dict[new_ticker.upper()] = new_full_name
# Sidebar for viewing historical trends
st.sidebar.header("View Historical Trends")
historical_ticker_input = st.sidebar.selectbox("Stock Ticker", [f"{key} - {value}" for key, value in stock_dict.items()], key="sidebar_historical_ticker")
if st.sidebar.button("View Historical Data"):
ticker = historical_ticker_input.split(' - ')[0]
data = display_historical_data(ticker)
st.sidebar.line_chart(data['Close'])
# Tabs for different functionalities
tab1, tab2, tab3 = st.tabs(["Today's Price", "Next Month's Price", "Historical Data"])
with tab1:
st.header("Today's Price")
ticker_input = st.selectbox("Stock Ticker", [f"{key} - {value}" for key, value in stock_dict.items()], key="today_ticker")
open_price = st.number_input("Open Price", value=0.0, key="today_open_price")
close_price = st.number_input("Close Price", value=0.0, key="today_close_price")
if st.button("Predict Today's Price"):
ticker = ticker_input.split(' - ')[0]
result = predict_stock_price(ticker, open_price, close_price)
st.write(result)
with tab2:
st.header("Next Month's Price")
next_month_ticker_input = st.selectbox("Stock Ticker", [f"{key} - {value}" for key, value in stock_dict.items()], key="next_month_ticker")
next_month_close_price = st.number_input("Close Price", value=0.0, key="next_month_close_price")
if st.button("Predict Next Month's Price"):
ticker = next_month_ticker_input.split(' - ')[0]
result = predict_next_month_price(ticker, next_month_close_price)
st.write(result)
with tab3:
st.header("Historical Data")
historical_ticker_input = st.selectbox("Stock Ticker", [f"{key} - {value}" for key, value in stock_dict.items()], key="historical_ticker")
if st.button("View Data"):
ticker = historical_ticker_input.split(' - ')[0]
data = display_historical_data(ticker)
st.dataframe(data)