MLflow 1.3 is out! It includes a bunch of bug fixes, performance enhancements and some new features. The official changelog can be found here:
Let’s have a look at some of the new features!
Tensorflow 2.0 support
A stable version of Tensorflow 2.0 was released just a few days ago, and now MLflow 1.3 has been released with Tensorflow 2.0 support. Check the official mlflow documentation for details
The mlflow.tensorflow module provides an API for logging and loading TensorFlow models. This module exports TensorFlow…
GeoJSON artifact previews
GeoJSON is a format for representing simple geographical features, like points, linestrings and polygons. MLflow 1.3 supports GeoJSON visualization in its tracking UI. This kind of visualization could be useful for example if you are making a machine learning model that denoises noisy linestrings of positions that come from mobile devices.
GeoJSON artifacts can be logged like this in MLflow:
The GeoJSON artifact preview looks like this in the UI:
HTML artifact previews
MLflow 1.3 lets you preview a HTML file artifact inside an iframe in the MLflow tracking UI. This could be useful for interactive plots, interactive visualizations of evolved 3D models or audio clips.
HTML artifacts can be logged like this in MLflow:
For example, if your run involved training an autoencoder, you could try to visualize the encoded vectors in a T-SNE plot, made interactive with sigma.js, like this:
I really wish I had this when I worked on my master’s thesis a couple of years ago!
Hosting a remote MLflow 1.3 tracking UI
If you need an MLflow 1.3 tracking UI with artifact storage in the cloud, check out MFlux.ai, which can set that up for individuals or teams at the click of a button.