After you have created an evaluation job, you can edit some of its configuration details. If you want to change configuration details that are not editable, you will need to delete the evaluation job and create a new one.
Before you begin
You must have already created an evaluation job.
View evaluation details
To update, pause, or delete an evaluation job, you must first navigate to the evaluation details page of the model version that has the job attached:
Open the AI Platform models page in the Cloud Console:
Click on the name of the model containing the model version with the evaluation job that you want to change.
Click on the name of the model version with the evaluation job that you want to change.
Click on the Evaluation tab.
Click on Evaluation details.
Pause the evaluation job
To pause the evaluation job, click the Pause button. While the job is paused, your model version's prediction input and output is not saved to the evaluation job's BigQuery table. The evaluation job doesn't run, so AI Platform Data Labeling Service doesn't send data to human reviewers for ground truth labeling. It also doesn't calculate evaluation metrics.
When you want continuous evaluation to start again, click the Resume button.
Update instructions or sampling percentage
If your evaluation job sends data to human reviewers for labeling, you can update the labeling instructions by uploading a new PDF file to Cloud Storage and providing the path to this file in the New instruction PDF (Optional) field. Learn how to write good instructions.
You can also update the percentage of your model version's predictions that get sampled and saved to your BigQuery table, as well as the maximum number of predictions to sample during each evaluation period.
If you change any of these settings, make sure to click Save afterward.
If you want to change anything else about the evaluation job, you must delete the job and create a new one with the configuration that you want.
Delete the evaluation job
To delete the evaluation job, click Delete, then click Continue.
Deleting an evaluation job doesn't delete the job's BigQuery table, so you can reuse this table if you create another job. It also doesn't delete any other Data Labeling Service resources (datasets, annotation specification sets, or instructions) created by the job, so you can continue to refer to these later.
Learn more about other Data Labeling Service features.