This page provides information about how to use Cloud Logging toview details about a Vertex AI model. UsingLogging, you see:

• The hyperparameters of the final model as key-value pairs.
• The hyperparameters and object values used during model training and tuning, as well as an objectivevalue.

By default, logs are deleted after 30 days.

The following topics are covered:

1. Viewing training logs.
2. Log fields.

Note: Model architecture logs are provided as part of theCloud Logging service. For general information about Cloud Logging, seetheCloud Logging documentation.

Before you begin

Before you can view the hyperparameter logs for your model, you musttrain it.

To perform this task, you must have the followingpermissions:

• logging.logServiceIndexes.list on the project
• logging.logServices.list on the project

Viewing training logs

You can use the Google Cloud console to access the hyperparameter logs of thefinal model and the hyperparameter logs of the tuning trials.

1. In the Google Cloud console, go to the Vertex AIModels page.

   Go to Models page

2. In theRegion drop-down, select the region where your model is located.

3. From the list of models, select your model.

4. Select your model's version number.

5. Open theVersion Details tab.

6. To see the hyperparameter log of the final model, go to theModel hyperparameters row and clickModel.

   1. There is just one log entry. Expand the payload as shown below.For details, seeLog fields.

      

7. To see the hyperparameter log of the tuning trials, go to theModel hyperparameters row and clickTrials.

   1. There is one entry for each of the tuning trials. Expand the payload asshown below. For details, seeLog fields.

      

Log fields

Activity logs are structured as described in theLogEntry typedocumentation.

Vertex AI model logs have, among other fields:

• labels: Thelog_type field is set toautoml_tables.
• jsonPayload: The specific details of the log entry, provided in JSONobject format. For details, seePayload contents for the hyperparameter log of the final modelorPayload contents for the hyperparameter log of a tuning trial.
• timestamp: The date and time when the model was created or the trial was run.

Payload contents for the hyperparameter log of the final model

ThejsonPayload field for the hyperparameter log of the final model contains amodelParameters field. This field contains one entry for each model thatcontributes to the final ensemble model. Each entry has ahyperparametersfield, whose contents depend on the model type. For details, seeList of hyperparameters.

Payload contents for the hyperparameter log of a tuning trial

ThejsonPayload field for the hyperparameter log of a tuning trial contains the following fields:

Field	Type	Description
modelStructure	JSON	A description of the Vertex AI model structure. This field contains amodelParameters field. ThemodelParameters field has ahyperparameters field, whose contents depend on the model type. For details, seeList of hyperparameters.
trainingObjectivePoint	JSON	The optimization objective used for model training. This entry includes a timestamp and an objective value at the time the log entry was recorded.

List of hyperparameters

The hyperparameter data provided in the logs differ for each type ofmodel. The following sections describe the hyperparameters for eachmodel type.

Gradient boosted decision tree models

• Tree L1 regularization
• Tree L2 regularization
• Max tree depth
• Model type:GBDT
• Number of trees
• Tree complexity

Feedforward neural network models

• Dropout rate
• Enable batchNorm (True orFalse)
• Enable embedding L1 (True orFalse)
• Enable embedding L2 (True orFalse)
• Enable L1 (True orFalse)
• Enable L2 (True orFalse)
• Enable layerNorm (True orFalse)
• Enable numerical embedding (True orFalse)
• Hidden layer size
• Model type:nn
• Normalize numerical column (True orFalse)
• Number of cross layers
• Number of hidden layers
• Skip connections type (dense,disable,concat, orslice_or_padding)

What's next

Once you're ready to make predictions with your classification or regressionmodel, you have two options:

• Make online (real-time) predictions using your model.
• Get batch predictions directly from your model.

Additionally, you can:

• Evaluate your model.
• Review general information about Cloud Logging.
• You can export your logs to BigQuery, Cloud Storage, orPub/Sub. ReadRoute logs to supported destinationsin the Logging documentation to learn how to export activity logs.