Hello tabular data: Deploy a model and request a prediction

After your AutoML tabular classification model is done training, create an endpoint and deploy your model to the endpoint. After your model is deployed to this new endpoint, test your model by requesting a prediction.

Load your model

When your model finishes training, it is listed in the Models tab.

1. In the Google Cloud console, in the Vertex AI section, go to the Models page.

   Go to the Models page

2. From the models list, click the name of your trained model that you created previously

3. Models are organized into versions. Click model version number 1.

Evaluate your model

The Evaluate panel helps you understand how the model performed against the test set. When you are done, continue to the next part of the tutorial.

Evaluation metrics

Optional. Hold the pointer over the ? icons to learn about each evaluation metric.

Optional. Move the confidence threshold slider to see how the precision, recall, and F1 scores are affected.

Confusion matrix

The confusion matrix shows how a prediction compares to the test set (ground truth).

Recall that label "1" is the negative class (the customer did not sign up for a term deposit) and "2" is the positive class. Your model likely did a better job predicting the negative class than the positive class. Perhaps with additional training time, more data, or additional features, you could improve predictive performance for the positive class.

Feature importance

Feature importance shows how each feature impacted model training: The higher the value, the more impactful.

Your model probably shows that duration (how long the most recent communication between the bank and customer lasted, in seconds) contributed heavily to the prediction outcome.

Deploy your model to an endpoint

To test a model or make online predictions, you need to deploy it to an endpoint.

1. Open the Deploy & Test panel.

2. Under Deploy your model, click Deploy to endpoint.

3. Enter Structured_AutoML_Tutorial for the endpoint name.

4. Click Continue.

5. Keep the minimum compute node at 1 and don't enter a maximum.

6. Select n1-standard-2 machine type.

7. Click Continue.

8. Turn off model monitoring for this endpoint.

9. Click Deploy. to create your endpoint and deploy your model to it.

   Model deployment takes around 5 minutes. When your endpoint is ready, proceed to the next part of the tutorial.

   Note: Make sure to undeploy the model later so you aren't charged for additional compute resources. We'll show you how to do that at the end of this tutorial.

Request a prediction

Now that your model is deployed to an endpoint, you can send prediction requests. Rather than send a request through the API or gcloud, you can test your model on this page.

1. In the Test your model section, you'll see a Value column that's pre-filled. You can use those values or enter new ones.

2. At the bottom of the section, press Predict.

   For this model, a prediction result of 1 represents a negative outcome—a deposit is not made at the bank. A prediction result of 2 represents a positive outcome—a deposit is made at the bank.

   Your model will return a confidence score, which is the model's level of certainty that the selected label is the correct one. The default value probably returned a high confidence score.

3. Optional. Try changing duration to a much higher value and press Predict again.

What's next

• To avoid incurring unexpected charges, follow the instructions in Clean up your project

• Learn more about model evaluation.

• Learn more about model predictions.