Today I decided to try out Google’s Vertex AI — a machine learning platform that makes it easier to train and deploy ML models.
In the last few weeks I’ve been writing code in Python to train models, but Vertex AI provides a low-code platform and a set of tools that simplify the deployment of ML models at scale.
Despite the promises of ease-of-use, I struggled to get Vertex AI to work on a small dataset (1k rows of data). Both training runs I tried went for 5-12 minutes before throwing up errors I couldn’t figure out.
Meanwhile, I was able to get a simple logistic regression model going in Python and trained locally in seconds. Here’s the performance of the model on fictional heart attack data.
Doing the same thing on Vertex AI should be quicker, and easier to deploy into a app. But it looks like I still have lots to figure out about the platform.