Stock Price Prediction

An application that explores the performance of different machine learning algorithims for prediction a given stocks price over time.

1. Overview
2. Quick Start
   • Running as a container
   • Running as a python application
3. Configuration
4. Artifacts & Reports

Overview

The application takes a given stock (as avaialble through yahoo finance's API), models and parameter tunes the given machine learning model before producing a plotly report on the models trianing and test set predictive performance. Disclaimer: For production grade predictions, far more focus would be given to better feature engineering, including the inclusion of company financials, news & social sentiment & other statistically similar stocks.

Curretnly the project supports the following machine learning algorthims:

• XGBoost
• LSTM Network (Tensorflow)

The project is not designed to create production grade predictions of a given stock price over time, rather, the project tries to explore the heuristic performance of different models on a given forecsting problem.

Quick start

Clone the project into a local directory.

git clone https://github.com/christian-nickerson/stock_price_prediction

The application can either be run a locally run python application OR as a docker container.

Running as a container

The run as a container, check docker is correctly installed and running (if not, please follow the docker guide here):

docker container run hello-world

If "Hello from Docker" is returned correctly, the the image can be built. To build the application image, run the following:

docker build . --no-cache -t stock_price_prediction:latest -f src/Dockerfile

Once built, the application can be run using the following:

docker run --rm -e STOCK_SYMBOL="TSLA" -e MODEL_NAME="xgboost" -e DATA_YEARS="5" -e PARAM_SAMPLES="50" stock_price_prediction:latest

Running as a python application (Linux & Mac)

To run as a local python applicaton, ensure Python 3.9 and PIP is isntalled:

python --version
pip --version

Create and activate a virtual environment:

python3 -m venv env
source env/bin/activate

Install the application dependencies:

pip install -r requirements.txt

To configure the application, create a local .env file with the following variables:

STOCK_SYMBOL="TSLA"
MODEL_NAME="xgboost"
DATA_YEARS="5"
PARAM_SAMPLES="50"

To run the application, using the configuration variables, run:

python src/main.py

Configuration

The application uses environment variables to configure itself. These are either passed with the docker run command or as a seperate .env that is immported at run time. A breakdown of the varaibles is as follows:

Variable	Decription
STOCK_SYMBOL	Stock symbol to run price prediction on (must match Yahoo Finance symbols)
MODEL_NAME	Name of model to train
DATA_YEARS	Number of years of data to train model - can cause high memory usage if a large number of years is used
PARAM_SAMPLES	Number of samples from paramter space to use for hyperparamter tuning. The greater the number, the more memory required and the longer the train time

Please be aware, the application is set to utilise as much compute resource as is available locally / provided to the container. Given the intensity of machine learning, this may cause compute and memeory pressure and potentially crash other applications running concurrently.

Artifacts and Reports

The application has the 3 main folders:

• src: source code for training models
• artifacts: pre-trained model artifact library
• reports: pre-trained model perfromance reports

The application, once a model has been trained, will save the model into the artifact library and save the displayed report in the reports folder. Any new model will overwrite an exisitng model if already contained in the artifact library. Warning: if running as a container, ensure the internal artifact and reports folder are mounted to a local directory at runtime to ensure the model and report are saved.

There are a number of pre-trained models as saved examples as part of the application in each of there folders.