Try Gemini 1.5 Pro, our most advanced multimodal model in Vertex AI, and see what you can build with a 1M token context window. Try Gemini 1.5 Pro, our most advanced multimodal model in Vertex AI, and see what you can build with a 1M token context window.

Make a prediction

You're now ready to use the test data in df_for_prediction to make a prediction request. The prediction request invokes your model to predict what species of penguin is represented by the penguin characteristics in each row in df_for_prediction.

Prepare prediction test data

Before you can use the test data to create predictions, you remove the species column. Because the species of penguin is what you're predicting, it can't be included in the test data used to create a prediction. After you remove the species column, you convert the data to a Python list because that's what the predict method takes as input. Run the following code convert your data to a Python list:

# Remove the species column
df_for_prediction.pop(LABEL_COLUMN)

# Convert data to a Python list
test_data_list = df_for_prediction.values.tolist()

(Optional) View test data

To help you understand the test data, you can run the following line of code to view it:

test_data_list

In each row, the respective values in each of the six columns refer to the following characteristics of one penguin:

Column	Penguin characteristic
0	`island` - The island where a species of penguin is found. The island value mapping is `0` for `Dream`, `1` for `Biscoe`, and `2` for `Torgersen`.
1	`culmen_length_mm` - The length of the ridge along the top of the bill of a penguin.
2	`culmen_depth_mm` - The height of the bill of a penguin.
3	`flipper_length_mm` - The length of the flipper-like wing of a penguin.
4	`body_mass_g` - The mass of the body of a pen.
5	`sex` - The sex of the penguin. `0` is `FEMALE` and `1` is `MALE`.

Send the prediction request

To create a prediction request, pass the Python list of test data you created to the endpoint's predict method.

The predict method evaluates the characteristics in each row and uses them to predict what kind of penguin they represent. Run the following code to create your predictions. The returned predictions contain a list of rows, where each row has three columns (Adelie Penguin (Pygoscelis adeliae) (column 1), Chinstrap penguin (Pygoscelis antarctica) (column 2), or Gentoo penguin (Pygoscelis papua) (column 3)).

# Get your predictions.
predictions = endpoint.predict(instances=test_data_list)

# View the predictions
predictions.predictions

Each column in a row contains a value, and the higher the value, the greater the confidence that the species of the penguin represented by that column is a correct prediction. For example, in the following sample prediction output row, the model uses the characteristics of the sample penguin data row to predict that the penguin is most likely of the Adelie Penguin (Pygoscelis adeliae) species. This is because the highest value, 0.732703805, is in the first column.

[0.732703805, 0.233752429, 0.0335437432]

In the following code, the NumPy argmax method returns the column for each row that contains the highest value. The highest value corresponds to the prediction that is most likely correct. The second line displays the array of predictions.

# Get the prediction for each set of input data.
species_predictions = np.argmax(predictions.predictions, axis=1)

# View the best prediction for the penguin characteristics in each row.
species_predictions

Each result in the species_predictions array predicts which penguin species the values in the corresponding row of test data corresponds to. For example, the first value is 0, which maps to the Adelie Penguin (Pygoscelis adeliae) species. This means that your model predicts that the species of a penguin with the characteristics in the first row of your test data is Adelie Penguin (Pygoscelis adeliae).

Clean up resources

Now that you're done, you can continue to use your notebook to explore and learn more about the resources you created and how they work.

Delete your resources

When you're ready, we recommend that you delete the Google Cloud resources you created during this tutorial so that you don't incur unnecessary charges. There are two ways to delete your resources:

Delete your project, which also deletes all the resources associated with your project. For more information, see Shutting down (deleting) projects.
Run code that deletes your training job (a CustomTrainingJob object), model (a Model object), endpoint (an Endpoint object), and Cloud Storage bucket. This option retains your project and any other resources you might have created that you don't explicitly delete with your code.

You must undeploy your model before you can delete it by passing force=True to the endpoint.delete method.

To retain your project and delete only resources you created during this tutorial, run the following code in your notebook:

import os

# Delete the training job
job.delete()

# Delete the endpoint and undeploy the model from it
endpoint.delete(force=True)

# Delete the model
model.delete()

# Delete the storage bucket and its contents
bucket.delete(force=True)

Delete your Vertex AI Workbench instance

You can keep your Vertex AI Workbench instance to use for future work. If you keep it, make sure you are aware of its cost. For more information, see Vertex AI Workbench pricing.

If you want to delete the Vertex AI Workbench instance, do the following:

In the Google Cloud console, go to the Vertex AI Workbench Instances page.

Go to Instances
Select your Vertex AI Workbench instance.
In the upper menu, click Delete.
In the Delete instance confirmation dialog, click Confirm. It takes a few minutes for the deletion to complete.

Train a model