Notes
🔍

Vertex AI AutoML

Google Cloud's answer to "I have data and a prediction problem, but no ML team" is Vertex AI AutoML: a managed service where you register a dataset, say what to predict, set a training budget, and get back a deployable model without writing any modelling code.

The shape of a job
Every supported data type follows the same four steps inside Vertex AI:

  • create a managed dataset from BigQuery or Cloud Storage and mark the target column or labels;
  • launch a training job with a budget expressed in node hours — the unit you actually pay for;
  • the service handles preprocessing, model search and ensembling internally, then reports evaluation metrics;
  • deploy to an endpoint for online prediction, or run batch prediction over a table.

What it covers
Tabular classification and regression, forecasting, image classification and object detection, plus video and text tasks — all behind one API. An AutoML model produces the same dataset, model and endpoint objects as a custom-trained one, so it lands in the same registry and reuses the same monitoring, feature-attribution and explainability tooling.

The trade you are making
You surrender control of the search space and visibility into the final architecture. In return you get managed infrastructure, strong tabular baselines, and no pipeline to maintain. The costs worth watching are node hours during training and a deployed endpoint that bills by the hour whether or not it serves a single request — the surprise on the first invoice of a team that treated a demo as free.

Vertex AI AutoML is Google Cloud's managed AutoML service: you register a dataset, name the target, set a budget in node hours, and the platform handles preprocessing, model search and ensembling before exposing the result as a deployable endpoint or batch prediction job. It covers tabular, forecasting, image, video and text tasks, and matters because its models share the registry, monitoring and explainability tooling used by custom-trained ones.

Some people cook from scratch; others order a meal kit and let someone else do the shopping and the chopping. Vertex AI AutoML is Google's meal kit for machine learning.

You upload your data, point at the thing you want predicted — which customers will cancel, what is in this photograph — and say how much computing time you are willing to pay for. Google's computers do the rest and hand back a working model you can plug into an app. You never see the recipe, and that is the trade: far less skill and effort needed, far less control over the result, and a bill that keeps ticking for as long as you leave the model switched on.