Manage models

This document shows you how to manage BigQuery ML models, including copying models and renaming models.

Required roles

To get the permissions that you need to read and create BigQuery models, ask your administrator to grant you the BigQuery Data Editor (roles/bigquery.dataEditor) IAM role on the project. For more information about granting roles, see Manage access to projects, folders, and organizations.

This predefined role contains the permissions required to read and create BigQuery models. To see the exact permissions that are required, expand the Required permissions section:

Required permissions

The following permissions are required to read and create BigQuery models:

  • To read information from models: bigquery.models.getData
  • To create models: bigquery.models.create

You might also be able to get these permissions with custom roles or other predefined roles.

Rename models

You cannot change the name of an existing model. If you need to change the model's name, follow the steps to copy the model. When you specify the destination in the copy operation, use the new model name.

Copy models

You can copy one or more models from a source dataset to a destination dataset by:

  • Using the Trusted Cloud console.
  • Using the bq command-line tool's bq cp command.
  • Calling the jobs.insert API method directly and configuring a copy job or by using the client libraries.

Limitations on copying models

Model copy jobs are subject to the following limitations:

  • When you copy a model, the name of the destination model must adhere to the same naming conventions as when you create a model.
  • Model copies are subject to BigQuery limits on copy jobs.
  • Copying a model is not supported by the Trusted Cloud console.
  • Copying multiple source models in a single command is not supported.
  • When you copy a model by using the CLI, the --destination_kms_key flag is not supported.

Copy a model

You can copy a model by:

  • Using the command-line tool's bq cp command
  • Calling the jobs.insert API method and configuring a copy job or by using the client libraries

To copy a model:

Console

The Trusted Cloud console does not support copying models.

bq

Issue the bq cp command. Optional flags:

  • -f or --force overwrites an existing model in the destination dataset and doesn't prompt you for confirmation.

  • -n or --no_clobber returns the following error message if the model exists in the destination dataset: '[PROJECT_ID]:[DATASET].[MODEL]' already exists, skipping.

    If -n is not specified, the default behavior is to prompt you to choose whether to replace the destination model.

If the source or destination dataset is in a project other than your default project, add the project ID to the dataset names in the following format: PROJECT_ID:DATASET.

bq cp -f -n PROJECT_ID:DATASET.SOURCE_MODEL PROJECT_ID:DATASET.DESTINATION_MODEL

Replace the following:

  • PROJECT_ID: your project ID.
  • DATASET: the name of the source or destination dataset.
  • SOURCE_MODEL: the model you're copying.
  • DESTINATION_MODEL: the name of the model in the destination dataset.

Examples:

Enter the following command to copy mydataset.mymodel to mydataset2. 

bq cp mydataset.mymodel mydataset2.mymodel

Enter the following command to copy mydataset.mymodel and to overwrite a destination model with the same name. The source dataset is in your default project. The destination dataset is in myotherproject. The -f shortcut is used to overwrite the destination model without a prompt. 

bq cp -f mydataset.mymodel myotherproject:myotherdataset.mymodel

Enter the following command to copy mydataset.mymodel and to return an error if the destination dataset contains a model with the same name. The source dataset is in your default project. The destination dataset is in myotherproject. The -n shortcut is used to prevent overwriting a model with the same name. 

bq cp -n mydataset.mymodel myotherproject:myotherdataset.mymodel

Enter the following command to copy mydataset.mymodel to mydataset2 and to rename the model mymodel2. Both datasets are in your default project. Both datasets were created in the asia-northeast1 region.

bq cp mydataset.mymodel mydataset2.mymodel2

API

To copy a model by using the API, call the bigquery.jobs.insert method and configure a copy job. Specify your location in the location property in the jobReference section of the job resource.

You must specify the following values in your job configuration:

"copy": {
      "sourceTable": {       // Required
        "projectId": string, // Required
        "datasetId": string, // Required
        "tableId": string    // Required
      },
      "destinationTable": {  // Required
        "projectId": string, // Required
        "datasetId": string, // Required
        "tableId": string    // Required
      },
      "createDisposition": string,  // Optional
      "writeDisposition": string,   // Optional
    },

Where:

  • sourceTable: provides information about the model to be copied.
  • destinationTable: provides information about the new model.
  • createDisposition: specifies whether to create the model if it doesn't exist.
  • writeDisposition: specifies whether to overwrite an existing model.

Encrypt models

For more information about using a customer-managed encryption key (CMEK) to encrypt a model, see Use CMEK to protect BigQuery ML models.

What's next