Summer Project
Task for TensorFlow Practice
◼ Objective
Use TensorFlow as main framework to realize a practical real time series forecasting
application and build your own deep learning models to predict the S&P 500 index of US stock
market by customizing input features (you can directly use any indicators created by using TA-
Lib).
◼ Due Date
January 29, 23:59, 2024
◼ Tasks
Predict the T+1 (the next trading date) close price of S&P 500
(https://hk.finance.yahoo.com/quote/%5EGSPC/ ) at time T (current trading date) by using
LSTM model, convolution neural network, or other specific NN models which are trained from
the historical data.
Key hints:
i. You can download historical time series data by using python package
yfinance: pip install yfinance
Document here:
This package encapsulates the web crawler as API that extracts data from yahoo
finance.
ii. Apply as much as possible models you can.
iii. The length of time series daily data should not shorter than 3 years.
iv. The input features should be generated at trading date T, the output prediction is the
CLOSE price at trading date T + 1.
v. You need to split the dataset into train, validation and test set at a reasonable ratio
and make sure these datasets are in chronological sequence.
vi. You should use Root Mean squared Error (RMSE) as metric.
◼ Results visualization
Visualize your results with different colors representing the historical and the prediction lines
by building a webpage as a demo.
You may write and establish a backend API server in flask or Django to hold the
models in memory, and call the API to do predicting, and then display your
prediction at front-end.
◼ Report:
A. Your models and the pipelines / structures.
B. How do you pre-process your dataset? Cleaning, augment, etc.
C. How to split your dataset?
D. What is your loss and metric for these models?
E. How to decide the hyperparameters?
F. What are the performances of your models?
G. Any ideas to upgrade your prediction?
