MLflow Demo + Model Review Overview

This repo contains materials for a talk on reviewing Machine Learning models before deploying to production, inspired by the concept of a Code Review. Most of the content is a tutorial on using MLflow Tracking.

See either the Jupyter Notebook (tutorial.ipynb) for the content, or the pdf slides which were converted from html rendered using RISE. The version of the talk on the main branch was presented at PyCon 2022. A previous version was presented at the PyData SLC Meetup in January 2022 (available under the pydata-slc-2022 branch), and another version was presented at the University of Utah's Data Science Seminar in September 2022 (see the uofu-ds-seminar branch). The title of the talks were "Model Review: improving transparency, reproducibility, & knowledge sharing using MLflow."

Getting Started

To try out the code examples in the notebook, you can create a conda environment from the yaml file (or install the dependencies listed there in your preferred way):

$ conda env create -f environment.yml

Then activate the environment

$ conda activate mlflow-demo

and try out some examples in the notebook. Enjoy!

Rendering as slides with RISE

Optionally, if you want to use the RISE extension on Jupyter Notebook, to view the notebook as slides, you might also need to run jupyter contrib nbextension install --user. Note that this extension currently only works in jupter notebook, not jupyterlab. After starting jupyter notebook, you should have a tab to configure Nbextensions including RISE (I like to set the theme to night, the transition to none and check the box to enable scroll since some of these slides include large screenshots). In the notebook tab, you should have a button that looks like a barchart to Enter/Exit RISE Slideshow.