Working of hyperparameters tuning – Vertex AI Custom Model Hyperparameter and Deployment

Running your training application several times with different values for your selected hyperparameters, all within the bounds you define, is how hyperparameter tuning works. Vertex AI remembers the outcomes of previous tests and uses that information to improve performance in new ones. Following the completion of the task, users will be provided with a summary of all trials and the optimal value configuration based on the criteria you set.

To tune hyperparameters, the Vertex AI must be in direct contact with the training program. The training application specifies all the information that your model requires. Users are expected to provide all the hyperparameters that need to be considered in the process of optimization and the outcome metrics such as accuracy, r2score, and so on, users wish optimize.

The Python package called cloudml-hypertune, aids in passing metrics to Vertex AI. More information about the cloudml-hypertune package is available in the following link:

https://pypi.org/project/cloudml-hypertune

Vertex AI Vizier

You may fine-tune hyperparameters in complex machine learning models with the aid of Vertex AI Vizier, a black-box optimization tool. It may be challenging and time-consuming to manually adjust ML models when they contain several distinct hyperparameters. By adjusting the hyperparameters, Vertex AI Vizier maximizes the output of the model.

Users need to provide study configuration for Vizier to optimize machine learning models. Study configuration includes metrics that need to be optimized (such as accuracy, r2 score, and so on) and Al hyperparameters that will influence the metrics.

Study configuration is implemented and is then called a Study. A trial employs a study’s aims (metrics) and input values (hyperparameters or parameters). A trial is a collection of inputs that produces a quantifiable output called measurement. A study continues until a specific number of trials is reached or is stopped. Importantly, Vertex AI Vizier proposes trial inputs but does not run them.

Hyperparameter tuning for custom training is a built-in feature that uses Vertex AI Vizier for training jobs.