mlflow-stock-predictions

Hello and welcome!

I will show you how you can try to predict stock market prices. We'll use mlflow tool. For more details, check official documentation: MlFlow documentation.

Mlflow is great MlOps tool which allows you to monitor and experiment with your machine learing models. You can also easly share your created model with others using conda or docker. Finally, you can make deployments and then getting predictions via API.

Keep in mind is not financial advice!

Setup environment

First clone my repo on your computer:

• git clone [email protected]:doxenix/mlflow-stock-predictions.git

We'll work locally, so make sure you already have installed all needed libraries. I recommend using conda environment.

• Create new conda environment: create --name myenv
• conda activate myenv
• pip install -r req.txt (if you got any error during installation try: pip install -r req.txt --user or try with sudo)

Step 1 - Train and Track the model

launch mlflow ui --backend-store-uri sqlite:///mlruns.db Got to http://127.0.0.1:5000

You can also run just: mlflow ui but without add sqlite database (or any other) you will not be able to store your models and push to production.

Then create your experiment. I called USD_PLN_daily.

Now, check params.py. You will find all parameters. Don't hesitate to change and experiment with them! Be aware, your created experiment name and ID should be the same like in params.py file.

Run: python pipeline.py and wait. Your model shoud start to train.

Great, you trained your model! You also noticed that you have mlruns and models folders. Start making experiments as many time as you want!

You can also explore your models in Mlflow. Check your result predictions on test data. The plot is available in your folders. You can also check it via Mlflow page.

Step 2 - Register model to production

Pick your best trained model and go to http://127.0.0.1:5000 again. Click on yor model and register it.

Then, go to Models tab, click on your model and push to production stage.

Step 3 - Deploy and make predictions

Open new bash terminal and run bash deploy.sh. Your model name inside this scrpit must be the same you used in Step 2.

This launches a gunicorn server serving at the localhost 127.0.0.1:5000. Now you can score locally on the deployed produciton model as a REST point.

From another terminal with your activated conda env, send a POST request with our fresh data:

• python run_prediction.py

Inside your terminal you will see predicted price.

Step 4 - Serving model as Docker image (testing phase!)

docker build -t mlflow_image -f Dockerfile .

After that you can try running your experiment:

mlflow run . --no-conda --param-list config=0

Bonus

SMA.py script requires oanda.cfg file with your API key. You have to create account on OANDA.

Check OANDA API Guide how to generate yor API key.

You will also need tpqoa library. For more details check: tpqoa