The ML.TRIAL_INFO function

This document describes theML.TRIAL_INFO function, which lets you displayinformation about trials from a model that useshyperparameter tuning.

You can use this function with models that supporthyperparameter tuning. For moreinformation, seeEnd-to-end user journeys for ML models.

Syntax

ML.TRIAL_INFO(MODEL `PROJECT_ID.DATASET.MODEL_NAME`)

Arguments

ML.TRIAL_INFO takes the following arguments:

• PROJECT_ID: your project ID.
• DATASET: the BigQuery dataset thatcontains the model.
• MODEL_NAME: The name of the model.

Output

ML.TRIAL_INFO returns one row per trial with the following columns:

• trial_id: anINT64 value that contains the ID assigned to each trial inthe approximate order of trial execution.trial_id values start from1.
• hyperparameters: aSTRUCT value that contains the hyperparameters used inthe trial.
• hparam_tuning_evaluation_metrics: aSTRUCT value that contains theevaluation metrics appropriate to the hyperparameter tuning objectivespecified by thehparam_tuning_objectives argumentin theCREATE MODEL statement. Metrics are calculated from the evaluationdata. For more information about the datasets used in hyperparameter tuning,seeData split.
• training_loss: aFLOAT64 value that contains the loss of the trial duringthe last iteration, calculated using the training data.
• eval_loss: aFLOAT64 value that contains the loss of the trial during thelast iteration, calculated using the evaluation data.

• status: aSTRING value that contains the final status of the trial.Possible values include the following:

  • SUCCEEDED: the trial succeeded.
  • FAILED: the trial failed.
  • INFEASIBLE: the trial was not run due to an invalid combination ofhyperparameters.

• error_message: aSTRING value that contains the error message that isreturned if the trial didn't succeed. For more information, seeError handling.

• is_optimal: aBOOL value that indicates whether the trial had the bestobjective value. If multiple trials are marked as optimal, then the trialwith the smallesttrial_id value is used as the default trial during modelserving.

Example

The following query retrieves information of all trials for the modelmydataset.mymodel in your default project:

SELECT*FROMML.TRIAL_INFO(MODEL`mydataset.mymodel`)

What's next

• For information about hyperparameter tuning, seeHyperparameter tuning overview.