Vertex AI Vizier

Create Vertex AI Vizier studies

Now, we will discuss how can you make API requests to Vertex AI Vizier by using Python.

Before you begin

1. If you are visiting the first time to Google Cloud then you need to create an account to evaluate the performance of your Google products in real-world scenarios. New customers will also earn $ 300 in free credit to run, test, and deploy their workloads.
    
2. Select or create a Google Cloud project in the Google Cloud console, on the project selector page. 
    
3. Enable the Vertex AI API.
    
4. Now, you need to create a service account:
   1. In the console, go to the Create service account page.
   2. Select your project.
   3. In the Service account name field, enter a name. The console fills in the Service account ID field based on this name.
   4. In the Service account description field, enter a description. For example, Service accounts for quick start.
   5. Click Create and continue.
   6. To provide access to your project, grant the following role(s) to your service account: Project > Owner.
   7. In the Select a role list, select a role. For additional roles, click add Add another role and add each additional role.
   8. Click Done to finish creating the service account. Do not close your browser window. You will use it in the next step.
       
5. Create a service account key:In the console, click the email address for the service account that you created.
   1. Click Keys.
   2. Click Add key, and then click Create new key.
   3. Click Create. A JSON key file is downloaded to your computer.
   4. Click Close.
       
6. Set the environment variable GOOGLE_APPLICATION_CREDENTIALS to the path of the JSON file that contains your service account key. This variable only applies to your current shell session, so if you open a new session, set the variable again.

• Example: Linux or macOS
• Example: Windows

 Install the Vertex AI SDK for Python.

Define constants

To define the constants, run the following command, replacing PROJECT_ID and REGION with your project ID and region. Create your own study name or you can also use the suggested values.

import json
import datetime
from google.cloud import aiplatform
REGION = "REGION"
PROJECT_ID = "PROJECT_ID"
# The following placeholder variables are automatically filled in.
STUDY_DISPLAY_NAME = '{}_study_{}'.format(PROJECT_ID.replace('-', ''), datetime.datetime.now().strftime('%Y%m%d_%H%M%S')) #@param {type: 'string'}
ENDPOINT = REGION + '-aiplatform.googleapis.com'
PARENT = 'projects/{}/locations/{}'.format(PROJECT_ID, REGION)

Construct API requests

The following command-line API requests are written in Python.

Create a study

A survey is a series of experiments or trials designed to assist in hyperparameters or parameter optimization.

In these examples, the goal is to find y = x ^ 2 where x is in the range [-10. 10]. This example uses a function that has only one parameter and is easy to calculate to show how to use Vertex AI Vizier.

param_x = {
    'parameter_id': 'x',
    'double_value_spec': {
        'min_value': -10.0,
        'max_value': 10.0
    }
}
metric_y = {
    'metric_id': 'y',
    'goal': 'MAXIMIZE'
}
study = {
    'display_name': STUDY_DISPLAY_NAME,
    'study_spec': {
      'algorithm': 'RANDOM_SEARCH',
      'parameters': [param_x],
      'metrics': [metric_y],
    }
}

To create the study by using your study configuration, send the following request through VizierServiceClient. Use the returned STUDY_NAME to query the study.

vizier_client = aiplatform.gapic.VizierServiceClient(client_options=dict(api_endpoint=ENDPOINT))
study = vizier_client.create_study(parent=PARENT, study=study)
STUDY_NAME = study.name

View your study

After your study is created, you can find the study in the Google Cloud console, in the Vertex AI section, and on the Experiments page.

 

Get a study

To get a study, send the following request.

vizier_client.get_study({'name': STUDY_NAME})

List studies

To list studies in a specific project and region, send the following request:

vizier_client.list_studies({'parent': PARENT})

Evaluate the results

As you receive trial proposals, evaluate each trial and record each result as a measurement.

For example, if the function you want to optimize is y = x^2, evaluate the function at the values ​​of x suggested in your research. If the recommended value is 0.1, the function gives y = 0.1 * 0.1 and gives 0.01.

Add a measurement

After evaluating trial suggestions and obtaining measurements, add the measurements to the trial.

Use the following commands to save measurements and send requests. In this example, replace RESULT with your measurements. If the function to optimize is y = x^2 and the suggested value for x is 0.1, the result is 0.01.

vizier_client.add_trial_measurement({
        'trial_name': TRIAL_ID,
        'measurement': {
            'metrics': [{'metric_id': 'y', 'value':RESULT }]
        }
    })

Complete a trial

After adding all the measurements of the trail, you need to send a command to complete the trail.

It means that when the trial is complete, you can send a command to complete the trial, or send a command to add the final measurement to complete the trial.

Without Final Measurement:

Send the following request to complete a trial without adding a final measurement:

vizier_client.complete_trial({
      'name': TRIAL_ID
  })

With Final Measurement:

To complete a trial and include a final measurement, use the following commands, replacing RESULT with the final measurement.

vizier_client.complete_trial({
      'name': TRIAL_ID
      'final_measurement': {
        'metrics': [{'metric_id': 'y', 'value': RESULT}]
      }
  })

List trials

To list trials in a specific study, send the following request:

vizier_client.list_trials({
    'parent': STUDY_NAME
})

After completing the pending tests, you can call suggestTrials to ask for more suggestions and repeat the test scoring process.

List optimal trials

The following example shows list_optimal_trials. It returns a Pareto optimal trial for a multipurpose study or an optimal trial for a single purpose study.

vizier_client.list_optimal_trials({
    'parent': STUDY_NAME
})

Frequently Asked Questions

What is a managed dataset in Vertex AI?

We recommend using Vertex AI-managed datasets to train your custom models, instead of writing your training application to ingest training data directly. Managed datasets offer the following benefits: Manage your datasets in a central location. Easily create labels and multiple annotation sets.

What is a feature store?

A feature store is a tool for storing commonly used features. When data scientists develop features for a machine learning model, those features can be added to the feature store. This makes those features available for reuse.

How good is Vertex AI?

Vertex AI is a great platform to build some templates for learning machines. We can add a lot of data ( more than 1000GB) and it's scalable. It's very fast to build it, lunch it and compare.

Conclusion

In this article, we discussed Vertex AI Vizier. We also discussed its working and how it is different from custom training .

Refer to our Guided Path on Coding Ninjas Studio to upskill yourself in Data Structures and Algorithms, Competitive Programming, JavaScript, System Design, and many more. If you want to test your competency in programming, you may check out the mock test series and participate in the contests available on Coding Ninjas Studio!

Nevertheless, you may consider our paid courses to give your career an edge over others!

Upvote this blog if you find it helpful and engaging!

Happy Learning!