End-to-End AutoML Tables is a solution for classification and regression that allows you to choose what to control and what to automate.
By default, Vertex AI searches for the optimal set of training hyperparameters. These hyperparameters include the model type and the model parameters. It then trains multiple models with multiple sets of hyperparameters and creates a single, final model from an ensemble of the top models.
By default, Vertex AI makes a conservative hardware choice (best for smaller datasets).
There are three options for customizing this workflow:
- Skip Architecture Search
- Override Search Space
- Configure Hardware
Skip Architecture Search
With this option enabled, you provide the full set of hyperparameters (N sets of hyperparameters for the top N models). Typically, these hyperparameters are an artefact from a previous architecture search.
Override Search Space
With this option enabled, you provide fixed values for a subset of the hyperparameters. Vertex AI searches for the optimal values of the remaining unfixed hyperparameters. This option is a good choice if you have a strong preference for the model type. Here, the model type is either neural networks or boosted trees.
With this option enabled, you can configure the machine types and the number of machines for training. This option is a good choice if you have a large dataset and want to optimize the machine hardware accordingly.