Créer une tâche de réglage des hyperparamètres pour un package Python

Crée une tâche de réglage des hyperparamètres pour un package Python à l'aide de la méthode create_hyperparameter_tuning_job.

En savoir plus

Pour obtenir une documentation détaillée incluant cet exemple de code, consultez les articles suivants :

Exemple de code

Java

Avant d'essayer cet exemple, suivez les instructions de configuration pour Java décrites dans le guide de démarrage rapide de Vertex AI à l'aide des bibliothèques clientes. Pour en savoir plus, consultez la documentation de référence de l'API Vertex AI Java.

Pour vous authentifier auprès de Vertex AI, configurez le service Identifiants par défaut de l'application. Pour en savoir plus, consultez Configurer l'authentification pour un environnement de développement local. 

import com.google.cloud.aiplatform.v1.AcceleratorType;
import com.google.cloud.aiplatform.v1.CustomJobSpec;
import com.google.cloud.aiplatform.v1.HyperparameterTuningJob;
import com.google.cloud.aiplatform.v1.JobServiceClient;
import com.google.cloud.aiplatform.v1.JobServiceSettings;
import com.google.cloud.aiplatform.v1.LocationName;
import com.google.cloud.aiplatform.v1.MachineSpec;
import com.google.cloud.aiplatform.v1.PythonPackageSpec;
import com.google.cloud.aiplatform.v1.StudySpec;
import com.google.cloud.aiplatform.v1.StudySpec.MetricSpec;
import com.google.cloud.aiplatform.v1.StudySpec.MetricSpec.GoalType;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec.ConditionalParameterSpec;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec.ConditionalParameterSpec.DiscreteValueCondition;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec.DiscreteValueSpec;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec.DoubleValueSpec;
import com.google.cloud.aiplatform.v1.StudySpec.ParameterSpec.ScaleType;
import com.google.cloud.aiplatform.v1.WorkerPoolSpec;
import java.io.IOException;
import java.util.Arrays;

public class CreateHyperparameterTuningJobPythonPackageSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace these variables before running the sample.
    String project = "PROJECT";
    String displayName = "DISPLAY_NAME";
    String executorImageUri = "EXECUTOR_IMAGE_URI";
    String packageUri = "PACKAGE_URI";
    String pythonModule = "PYTHON_MODULE";
    createHyperparameterTuningJobPythonPackageSample(
        project, displayName, executorImageUri, packageUri, pythonModule);
  }

  static void createHyperparameterTuningJobPythonPackageSample(
      String project,
      String displayName,
      String executorImageUri,
      String packageUri,
      String pythonModule)
      throws IOException {
    JobServiceSettings settings =
        JobServiceSettings.newBuilder()
            .setEndpoint("us-central1-aiplatform.googleapis.com:443")
            .build();
    String location = "us-central1";

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests. After completing all of your requests, call
    // the "close" method on the client to safely clean up any remaining background resources.
    try (JobServiceClient client = JobServiceClient.create(settings)) {
      // study spec
      MetricSpec metric =
          MetricSpec.newBuilder().setMetricId("val_rmse").setGoal(GoalType.MINIMIZE).build();

      // decay
      DoubleValueSpec doubleValueSpec =
          DoubleValueSpec.newBuilder().setMinValue(1e-07).setMaxValue(1).build();
      ParameterSpec parameterDecaySpec =
          ParameterSpec.newBuilder()
              .setParameterId("decay")
              .setDoubleValueSpec(doubleValueSpec)
              .setScaleType(ScaleType.UNIT_LINEAR_SCALE)
              .build();
      Double[] decayValues = {32.0, 64.0};
      DiscreteValueCondition discreteValueDecay =
          DiscreteValueCondition.newBuilder().addAllValues(Arrays.asList(decayValues)).build();
      ConditionalParameterSpec conditionalParameterDecay =
          ConditionalParameterSpec.newBuilder()
              .setParameterSpec(parameterDecaySpec)
              .setParentDiscreteValues(discreteValueDecay)
              .build();

      // learning rate
      ParameterSpec parameterLearningSpec =
          ParameterSpec.newBuilder()
              .setParameterId("learning_rate")
              .setDoubleValueSpec(doubleValueSpec) // Use the same min/max as for decay
              .setScaleType(ScaleType.UNIT_LINEAR_SCALE)
              .build();

      Double[] learningRateValues = {4.0, 8.0, 16.0};
      DiscreteValueCondition discreteValueLearning =
          DiscreteValueCondition.newBuilder()
              .addAllValues(Arrays.asList(learningRateValues))
              .build();
      ConditionalParameterSpec conditionalParameterLearning =
          ConditionalParameterSpec.newBuilder()
              .setParameterSpec(parameterLearningSpec)
              .setParentDiscreteValues(discreteValueLearning)
              .build();

      // batch size
      Double[] batchSizeValues = {4.0, 8.0, 16.0, 32.0, 64.0, 128.0};

      DiscreteValueSpec discreteValueSpec =
          DiscreteValueSpec.newBuilder().addAllValues(Arrays.asList(batchSizeValues)).build();
      ParameterSpec parameter =
          ParameterSpec.newBuilder()
              .setParameterId("batch_size")
              .setDiscreteValueSpec(discreteValueSpec)
              .setScaleType(ScaleType.UNIT_LINEAR_SCALE)
              .addConditionalParameterSpecs(conditionalParameterDecay)
              .addConditionalParameterSpecs(conditionalParameterLearning)
              .build();

      // trial_job_spec
      MachineSpec machineSpec =
          MachineSpec.newBuilder()
              .setMachineType("n1-standard-4")
              .setAcceleratorType(AcceleratorType.NVIDIA_TESLA_T4)
              .setAcceleratorCount(1)
              .build();

      PythonPackageSpec pythonPackageSpec =
          PythonPackageSpec.newBuilder()
              .setExecutorImageUri(executorImageUri)
              .addPackageUris(packageUri)
              .setPythonModule(pythonModule)
              .build();

      WorkerPoolSpec workerPoolSpec =
          WorkerPoolSpec.newBuilder()
              .setMachineSpec(machineSpec)
              .setReplicaCount(1)
              .setPythonPackageSpec(pythonPackageSpec)
              .build();

      StudySpec studySpec =
          StudySpec.newBuilder()
              .addMetrics(metric)
              .addParameters(parameter)
              .setAlgorithm(StudySpec.Algorithm.RANDOM_SEARCH)
              .build();
      CustomJobSpec trialJobSpec =
          CustomJobSpec.newBuilder().addWorkerPoolSpecs(workerPoolSpec).build();
      // hyperparameter_tuning_job
      HyperparameterTuningJob hyperparameterTuningJob =
          HyperparameterTuningJob.newBuilder()
              .setDisplayName(displayName)
              .setMaxTrialCount(4)
              .setParallelTrialCount(2)
              .setStudySpec(studySpec)
              .setTrialJobSpec(trialJobSpec)
              .build();
      LocationName parent = LocationName.of(project, location);
      HyperparameterTuningJob response =
          client.createHyperparameterTuningJob(parent, hyperparameterTuningJob);
      System.out.format("response: %s\n", response);
      System.out.format("Name: %s\n", response.getName());
    }
  }
}

Python

Avant d'essayer cet exemple, suivez les instructions de configuration pour Python décrites dans le guide de démarrage rapide de Vertex AI à l'aide des bibliothèques clientes. Pour en savoir plus, consultez la documentation de référence de l'API Vertex AI Python.

Pour vous authentifier auprès de Vertex AI, configurez le service Identifiants par défaut de l'application. Pour en savoir plus, consultez Configurer l'authentification pour un environnement de développement local. 

from google.cloud import aiplatform


def create_hyperparameter_tuning_job_python_package_sample(
    project: str,
    display_name: str,
    executor_image_uri: str,
    package_uri: str,
    python_module: str,
    location: str = "us-central1",
    api_endpoint: str = "us-central1-aiplatform.googleapis.com",
):
    # The AI Platform services require regional API endpoints.
    client_options = {"api_endpoint": api_endpoint}
    # Initialize client that will be used to create and send requests.
    # This client only needs to be created once, and can be reused for multiple requests.
    client = aiplatform.gapic.JobServiceClient(client_options=client_options)

    # study_spec
    metric = {
        "metric_id": "val_rmse",
        "goal": aiplatform.gapic.StudySpec.MetricSpec.GoalType.MINIMIZE,
    }

    conditional_parameter_decay = {
        "parameter_spec": {
            "parameter_id": "decay",
            "double_value_spec": {"min_value": 1e-07, "max_value": 1},
            "scale_type": aiplatform.gapic.StudySpec.ParameterSpec.ScaleType.UNIT_LINEAR_SCALE,
        },
        "parent_discrete_values": {"values": [32, 64]},
    }
    conditional_parameter_learning_rate = {
        "parameter_spec": {
            "parameter_id": "learning_rate",
            "double_value_spec": {"min_value": 1e-07, "max_value": 1},
            "scale_type": aiplatform.gapic.StudySpec.ParameterSpec.ScaleType.UNIT_LINEAR_SCALE,
        },
        "parent_discrete_values": {"values": [4, 8, 16]},
    }
    parameter = {
        "parameter_id": "batch_size",
        "discrete_value_spec": {"values": [4, 8, 16, 32, 64, 128]},
        "scale_type": aiplatform.gapic.StudySpec.ParameterSpec.ScaleType.UNIT_LINEAR_SCALE,
        "conditional_parameter_specs": [
            conditional_parameter_decay,
            conditional_parameter_learning_rate,
        ],
    }

    # trial_job_spec
    machine_spec = {
        "machine_type": "n1-standard-4",
        "accelerator_type": aiplatform.gapic.AcceleratorType.NVIDIA_TESLA_K80,
        "accelerator_count": 1,
    }
    worker_pool_spec = {
        "machine_spec": machine_spec,
        "replica_count": 1,
        "python_package_spec": {
            "executor_image_uri": executor_image_uri,
            "package_uris": [package_uri],
            "python_module": python_module,
            "args": [],
        },
    }

    # hyperparameter_tuning_job
    hyperparameter_tuning_job = {
        "display_name": display_name,
        "max_trial_count": 4,
        "parallel_trial_count": 2,
        "study_spec": {
            "metrics": [metric],
            "parameters": [parameter],
            "algorithm": aiplatform.gapic.StudySpec.Algorithm.RANDOM_SEARCH,
        },
        "trial_job_spec": {"worker_pool_specs": [worker_pool_spec]},
    }
    parent = f"projects/{project}/locations/{location}"
    response = client.create_hyperparameter_tuning_job(
        parent=parent, hyperparameter_tuning_job=hyperparameter_tuning_job
    )
    print("response:", response)

Étapes suivantes

Pour rechercher et filtrer des exemples de code pour d'autres produits Google Cloud, consultez l'explorateur d'exemples Google Cloud.