This project includes the course project and study logs done during the course.

This project is a demo of a Django server which pulls trading data such as EUR/USD prices using the FXCM API and allows users to create models to run on the data. The models have an option to be either "test" or "production" models, where the idea is that the production models would actually trade.

The project is meant to be used, and is built for the Heroku platform, where various of the subtasks are partitioned to their own subtasks. Heroku would allow for the dynamic changing of the subprocesses processing capabilities. Heroku would also offer free SSL security and allow users to access the platform on the web.

Clone repository

git clone git@github.com:jonatanvm/s4ml-project.git

cd s4ml-project

Start and initialize new virtual environment

virutualenv venv

source venv/bin/activate

Install requirements (and update pip)

pip install --upgrade pip pip install -r requirements.txt

The easiest way to run the project by using the Heroku CLI.

If you have it installed simply create and .env file in the root folder with the contents:

SECRET_KEY=YOURSECRETKEY

Then you can just run the project with:

heroku local

If you don't have Heroku CLI you need to run all the subprocesses individually:

export DJANGO_SETTINGS_MODULE='s4project.settings'
export SECRET_KET=YOURSECRET
python manage.py runserver
python get_price_data.py
python run_models.py

If you run get_price_data.py and run_models.py in different terminals you need to run export DJANGO_SETTINGS_MODULE='s4project.settings' in each terminal.

On the main page click Add model.

The attached python script must have two methods: run_test and predict

run_test takes an array of the most recent data as a parameter and must return a tuple rmse, mae, r2, profit, accuracy, parameters , where all the values are floats except paremeters which is a dictionary of the parameters of the model.

predict(X, y, p) takes the price data y, the time data X and value to predict p as arguments and must return the tuple prediction, parameters where prediction is the predicted value and parameters is a dict of parameters used in the model.