The purpose of the tutorial is to give a taste of the ML development process with MLflow. It is intended in providing a breadth rather than depth introduction but will give the audience a hands-on experience allowing them to dive into the specific submodules.
You will need a linux OS or a macOS with docker and visual studio code installed. I have not tried this setup on windows. If you are using windows the recommended way is to use github codespaces.
Please come to the tutorial already with a running repository. Setting up the environment can take some time and we would like to spend the time in the tutorial coding.
If you already have VS Code and Docker installed, you can click the badge above. Clicking these links will cause VS Code to automatically install the Dev Containers extension if needed, clone the source code into a container volume, and spin up a dev container for use.
If you are using windows or do not want to install vscode on your machine, you can click on the badge above to spin up a codespace environment.
If you want to do it manually you will need to create a new environemnent with python 3.10 installed (you will need to change some path on the top of the notebook).
Then you can install the requirements using pip
pip install -r requirements.txt -r requirements-dev.txtYou would also need to have java and pyenv installed to run the predict step in the second notebook.
To make sure your environement was set up correctly you can finally run the pipeline end to end.
python src/scripts/run_all.py localWe will have 2 git branchs you can switch between:
mainbranch which will contain the code with the solutiontutorialbranch which contains the code with the ToDo to implement.
If you got the code from clicking on the codespace link you are already in the tutorial branch. Otherwise you can switch between from main to solution by running:
git checkout tutorialIf you are stuck you can check the solution by going to the main branch.
The tutorial consists of 2 jupyter notebooks which can be found in src/notebook.
You can start the jupyter server by running jupyter lab or use your IDE directly (pycharm pro, vscode and codespace supports jupyter notebooks).
The presentation for the workshop can be found here.
This repository was only possible thanks to others repository that as used. In particular I would like to call out: