MLflow

MLflow is a tool that keeps a log of the process of building machine learning models, like a diary. To use a cooking analogy, it's like a notebook where you jot down ingredient ratios and taste ratings every time you try a recipe dozens of different ways. It lets you look back later and compare, at a glance, which settings you trained with, how well each one performed, and which version turned out best.

Finishing a single model often involves retraining it hundreds of times while tweaking data and settings, and it's hard for a person to keep track of which attempt worked best. MLflow automatically records these experiment logs, and it also manages finished models so they can be saved and later connected to a live service. It was created by Databricks and is a widely used open-source tool that many companies and developers freely adopt.

To get a feel for MLflow yourself, try the following in an environment with Python installed:

1. Install it with the command pip install mlflow.
2. Add two lines of code—one to start and one to end experiment tracking—around the model training code you normally use.
3. After training finishes, visit the address shown on screen, and you'll see a table summarizing the settings and results of the run you just did.
4. Run the same code two or three more times with different settings, and check the screen where the results are compared side by side.