训练视频分类模型

本页面介绍了如何使用 Google Cloud 控制台或 Vertex AI API 根据视频数据集训练 AutoML 分类模型。

训练 AutoML 模型

Google Cloud 控制台

在 Google Cloud 控制台的 Vertex AI 部分中，前往数据集页面。

转到“数据集”页面
点击要用于训练模型的数据集的名称，以打开其详情页面。
点击训练新模型。
输入新模型的显示名。
如果您要手动设置训练数据的拆分方式，请展开高级选项，然后选择数据拆分选项。了解详情。
点击继续。
选择模型训练方法。
- AutoML 非常适合各种用例。
- Seq2seq+ 非常适合进行实验。该算法可能比 AutoML 收敛更快，因为它的架构更简单，并且使用较小的搜索空间。我们的实验发现，在时间预算较少的情况下以及在小于 1 GB 的数据集上，Seq2Seq+ 性能较好。
点击继续。
点击开始训练。
模型训练可能需要几个小时，具体取决于数据的大小和复杂性，以及训练预算（如果指定）。您可以关闭此标签页，稍后再返回。模型完成训练后，您会收到电子邮件。

训练开始几分钟后，您即可在模型的属性信息中查看训练节点时的估算值。如果您取消训练，当前产品不会产生费用。

API

在下面选择您的语言或环境对应的标签页：

REST

在使用任何请求数据之前，请先进行以下替换：

LOCATION：数据集所在且用于存储模型的区域。例如 us-central1。
PROJECT：您的项目 ID。
MODEL_DISPLAY_NAME：新训练模型的显示名称。
DATASET_ID：训练数据集的 ID。
filterSplit 对象为可选；您可使用它来控制数据拆分。如需详细了解如何控制数据拆分，请参阅使用 REST 控制数据拆分。
PROJECT_NUMBER：您项目的自动生成的项目编号

HTTP 方法和网址：

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT/locations/LOCATION/trainingPipelines

请求 JSON 正文：

{
    "displayName": "MODE_DISPLAY_NAME",
    "trainingTaskDefinition": "gs://google-cloud-aiplatform/schema/trainingjob/definition/automl_video_classification_1.0.0.yaml",
    "trainingTaskInputs": {},
    "modelToUpload": {"displayName": "MODE_DISPLAY_NAME"},
    "inputDataConfig": {
      "datasetId": "DATASET_ID",
      "filterSplit": {
        "trainingFilter": "labels.ml_use = training",
        "validationFilter": "labels.ml_use = -",
        "testFilter": "labels.ml_use = test"
      }
    }
}

如需发送您的请求，请展开以下选项之一：

curl（Linux、macOS 或 Cloud Shell）

注意：以下命令假定您已使用您的用户账号通过运行 gcloud init 或 gcloud auth login 登录 gcloud CLI，或者使用了 Cloud Shell，这会使您自动登录 gcloud CLI。您可以运行 gcloud auth list 来检查当前活跃的账号。

将请求正文保存在名为 request.json 的文件中，然后执行以下命令：

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT/locations/LOCATION/trainingPipelines"

PowerShell (Windows)

注意：以下命令假定您已使用您的用户账号通过运行 gcloud init 或 gcloud auth login 登录 gcloud CLI。您可以运行 gcloud auth list 来检查当前活跃的账号。

将请求正文保存在名为 request.json 的文件中，然后执行以下命令：

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT/locations/LOCATION/trainingPipelines" | Select-Object -Expand Content

您应该收到类似以下内容的 JSON 响应：

{
  "name": "projects/PROJECT_NUMBER/locations/us-central1/trainingPipelines/2307109646608891904",
  "displayName": "myModelName",
  "trainingTaskDefinition": "gs://google-cloud-aiplatform/schema/trainingjob/definition/automl_video_classification_1.0.0.yaml",
  "modelToUpload": {
    "displayName": "myModelName"
  },
  "state": "PIPELINE_STATE_PENDING",
  "createTime": "2020-04-18T01:22:57.479336Z",
  "updateTime": "2020-04-18T01:22:57.479336Z"
}

Java

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Java 设置说明执行操作。如需了解详情，请参阅 Vertex AI Java API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭据。如需了解详情，请参阅为本地开发环境设置身份验证。


import com.google.cloud.aiplatform.util.ValueConverter;
import com.google.cloud.aiplatform.v1.FilterSplit;
import com.google.cloud.aiplatform.v1.FractionSplit;
import com.google.cloud.aiplatform.v1.InputDataConfig;
import com.google.cloud.aiplatform.v1.LocationName;
import com.google.cloud.aiplatform.v1.Model;
import com.google.cloud.aiplatform.v1.PipelineServiceClient;
import com.google.cloud.aiplatform.v1.PipelineServiceSettings;
import com.google.cloud.aiplatform.v1.PredefinedSplit;
import com.google.cloud.aiplatform.v1.TimestampSplit;
import com.google.cloud.aiplatform.v1.TrainingPipeline;
import com.google.rpc.Status;
import java.io.IOException;

public class CreateTrainingPipelineVideoClassificationSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace these variables before running the sample.
    String videoClassificationDisplayName =
        "YOUR_TRAINING_PIPELINE_VIDEO_CLASSIFICATION_DISPLAY_NAME";
    String datasetId = "YOUR_DATASET_ID";
    String modelDisplayName = "YOUR_MODEL_DISPLAY_NAME";
    String project = "YOUR_PROJECT_ID";
    createTrainingPipelineVideoClassification(
        videoClassificationDisplayName, datasetId, modelDisplayName, project);
  }

  static void createTrainingPipelineVideoClassification(
      String videoClassificationDisplayName,
      String datasetId,
      String modelDisplayName,
      String project)
      throws IOException {
    PipelineServiceSettings pipelineServiceSettings =
        PipelineServiceSettings.newBuilder()
            .setEndpoint("us-central1-aiplatform.googleapis.com:443")
            .build();

    // 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 (PipelineServiceClient pipelineServiceClient =
        PipelineServiceClient.create(pipelineServiceSettings)) {
      String location = "us-central1";
      LocationName locationName = LocationName.of(project, location);
      String trainingTaskDefinition =
          "gs://google-cloud-aiplatform/schema/trainingjob/definition/"
              + "automl_video_classification_1.0.0.yaml";

      InputDataConfig inputDataConfig =
          InputDataConfig.newBuilder().setDatasetId(datasetId).build();
      Model model = Model.newBuilder().setDisplayName(modelDisplayName).build();

      TrainingPipeline trainingPipeline =
          TrainingPipeline.newBuilder()
              .setDisplayName(videoClassificationDisplayName)
              .setTrainingTaskDefinition(trainingTaskDefinition)
              .setTrainingTaskInputs(ValueConverter.EMPTY_VALUE)
              .setInputDataConfig(inputDataConfig)
              .setModelToUpload(model)
              .build();

      TrainingPipeline trainingPipelineResponse =
          pipelineServiceClient.createTrainingPipeline(locationName, trainingPipeline);

      System.out.println("Create Training Pipeline Video Classification Response");
      System.out.format("\tName: %s\n", trainingPipelineResponse.getName());
      System.out.format("\tDisplay Name: %s\n", trainingPipelineResponse.getDisplayName());
      System.out.format(
          "\tTraining Task Definition: %s\n", trainingPipelineResponse.getTrainingTaskDefinition());
      System.out.format(
          "\tTraining Task Inputs: %s\n", trainingPipelineResponse.getTrainingTaskInputs());
      System.out.format(
          "\tTraining Task Metadata: %s\n", trainingPipelineResponse.getTrainingTaskMetadata());
      System.out.format("\tState: %s\n", trainingPipelineResponse.getState());
      System.out.format("\tCreate Time: %s\n", trainingPipelineResponse.getCreateTime());
      System.out.format("\tStart Time: %s\n", trainingPipelineResponse.getStartTime());
      System.out.format("\tEnd Time: %s\n", trainingPipelineResponse.getEndTime());
      System.out.format("\tUpdate Time: %s\n", trainingPipelineResponse.getUpdateTime());
      System.out.format("\tLabels: %s\n", trainingPipelineResponse.getLabelsMap());

      InputDataConfig inputDataConfigResponse = trainingPipelineResponse.getInputDataConfig();
      System.out.println("\tInput Data Config");
      System.out.format("\t\tDataset Id: %s\n", inputDataConfigResponse.getDatasetId());
      System.out.format(
          "\t\tAnnotations Filter: %s\n", inputDataConfigResponse.getAnnotationsFilter());

      FractionSplit fractionSplit = inputDataConfigResponse.getFractionSplit();
      System.out.println("\t\tFraction Split");
      System.out.format("\t\t\tTraining Fraction: %s\n", fractionSplit.getTrainingFraction());
      System.out.format("\t\t\tValidation Fraction: %s\n", fractionSplit.getValidationFraction());
      System.out.format("\t\t\tTest Fraction: %s\n", fractionSplit.getTestFraction());

      FilterSplit filterSplit = inputDataConfigResponse.getFilterSplit();
      System.out.println("\t\tFilter Split");
      System.out.format("\t\t\tTraining Fraction: %s\n", filterSplit.getTrainingFilter());
      System.out.format("\t\t\tValidation Fraction: %s\n", filterSplit.getValidationFilter());
      System.out.format("\t\t\tTest Fraction: %s\n", filterSplit.getTestFilter());

      PredefinedSplit predefinedSplit = inputDataConfigResponse.getPredefinedSplit();
      System.out.println("\t\tPredefined Split");
      System.out.format("\t\t\tKey: %s\n", predefinedSplit.getKey());

      TimestampSplit timestampSplit = inputDataConfigResponse.getTimestampSplit();
      System.out.println("\t\tTimestamp Split");
      System.out.format("\t\t\tTraining Fraction: %s\n", timestampSplit.getTrainingFraction());
      System.out.format("\t\t\tValidation Fraction: %s\n", timestampSplit.getValidationFraction());
      System.out.format("\t\t\tTest Fraction: %s\n", timestampSplit.getTestFraction());
      System.out.format("\t\t\tKey: %s\n", timestampSplit.getKey());

      Model modelResponse = trainingPipelineResponse.getModelToUpload();
      System.out.println("\tModel To Upload");
      System.out.format("\t\tName: %s\n", modelResponse.getName());
      System.out.format("\t\tDisplay Name: %s\n", modelResponse.getDisplayName());
      System.out.format("\t\tDescription: %s\n", modelResponse.getDescription());
      System.out.format("\t\tMetadata Schema Uri: %s\n", modelResponse.getMetadataSchemaUri());
      System.out.format("\t\tMeta Data: %s\n", modelResponse.getMetadata());
      System.out.format("\t\tTraining Pipeline: %s\n", modelResponse.getTrainingPipeline());
      System.out.format("\t\tArtifact Uri: %s\n", modelResponse.getArtifactUri());
      System.out.format(
          "\t\tSupported Deployment Resources Types: %s\n",
          modelResponse.getSupportedDeploymentResourcesTypesList().toString());
      System.out.format(
          "\t\tSupported Input Storage Formats: %s\n",
          modelResponse.getSupportedInputStorageFormatsList().toString());
      System.out.format(
          "\t\tSupported Output Storage Formats: %s\n",
          modelResponse.getSupportedOutputStorageFormatsList().toString());
      System.out.format("\t\tCreate Time: %s\n", modelResponse.getCreateTime());
      System.out.format("\t\tUpdate Time: %s\n", modelResponse.getUpdateTime());
      System.out.format("\t\tLables: %s\n", modelResponse.getLabelsMap());

      Status status = trainingPipelineResponse.getError();
      System.out.println("\tError");
      System.out.format("\t\tCode: %s\n", status.getCode());
      System.out.format("\t\tMessage: %s\n", status.getMessage());
    }
  }
}

Node.js

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Node.js 设置说明执行操作。如需了解详情，请参阅 Vertex AI Node.js API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭据。如需了解详情，请参阅为本地开发环境设置身份验证。

/**
 * TODO(developer): Uncomment these variables before running the sample.\
 * (Not necessary if passing values as arguments)
 */

// const datasetId = 'YOUR_DATASET_ID';
// const modelDisplayName = 'YOUR_MODEL_DISPLAY_NAME';
// const trainingPipelineDisplayName = 'YOUR_TRAINING_PIPELINE_DISPLAY_NAME';
// const project = 'YOUR_PROJECT_ID';
// const location = 'YOUR_PROJECT_LOCATION';
const aiplatform = require('@google-cloud/aiplatform');
const {definition} =
  aiplatform.protos.google.cloud.aiplatform.v1.schema.trainingjob;

// Imports the Google Cloud Pipeline Service Client library
const {PipelineServiceClient} = aiplatform.v1;

// Specifies the location of the api endpoint
const clientOptions = {
  apiEndpoint: 'us-central1-aiplatform.googleapis.com',
};

// Instantiates a client
const pipelineServiceClient = new PipelineServiceClient(clientOptions);

async function createTrainingPipelineVideoClassification() {
  // Configure the parent resource
  const parent = `projects/${project}/locations/${location}`;
  // Values should match the input expected by your model.
  const trainingTaskInputObj = new definition.AutoMlVideoClassificationInputs(
    {}
  );
  const trainingTaskInputs = trainingTaskInputObj.toValue();

  const modelToUpload = {displayName: modelDisplayName};
  const inputDataConfig = {datasetId: datasetId};
  const trainingPipeline = {
    displayName: trainingPipelineDisplayName,
    trainingTaskDefinition:
      'gs://google-cloud-aiplatform/schema/trainingjob/definition/automl_video_classification_1.0.0.yaml',
    trainingTaskInputs,
    inputDataConfig,
    modelToUpload,
  };
  const request = {
    parent,
    trainingPipeline,
  };

  // Create training pipeline request
  const [response] =
    await pipelineServiceClient.createTrainingPipeline(request);

  console.log('Create training pipeline video classification response');
  console.log(`Name : ${response.name}`);
  console.log('Raw response:');
  console.log(JSON.stringify(response, null, 2));
}
createTrainingPipelineVideoClassification();

Python

如需了解如何安装或更新 Vertex AI SDK for Python，请参阅安装 Vertex AI SDK for Python。如需了解详情，请参阅 Python API 参考文档。

from google.cloud import aiplatform
from google.cloud.aiplatform.gapic.schema import trainingjob


def create_training_pipeline_video_classification_sample(
    project: str,
    display_name: str,
    dataset_id: str,
    model_display_name: 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.PipelineServiceClient(client_options=client_options)
    training_task_inputs = (
        trainingjob.definition.AutoMlVideoClassificationInputs().to_value()
    )

    training_pipeline = {
        "display_name": display_name,
        "training_task_definition": "gs://google-cloud-aiplatform/schema/trainingjob/definition/automl_video_classification_1.0.0.yaml",
        # Training task inputs are empty for video classification
        "training_task_inputs": training_task_inputs,
        "input_data_config": {"dataset_id": dataset_id},
        "model_to_upload": {"display_name": model_display_name},
    }
    parent = f"projects/{project}/locations/{location}"
    response = client.create_training_pipeline(
        parent=parent, training_pipeline=training_pipeline
    )
    print("response:", response)

使用 REST 控制数据拆分

您可以控制在训练集、验证集和测试集之间拆分训练数据的方式。使用 Vertex AI API 时，请使用 Split 对象来确定数据拆分。Split 对象可以包含在 InputConfig 对象中作为多种对象类型中的一种，其中每种类型都提供一种不同的训练数据拆分方式。您只能选择一种方法。

FractionSplit:
- TRAINING_FRACTION：要用于训练集的训练数据的比例。
- VALIDATION_FRACTION：要用于验证集的训练数据的比例。不用于视频数据。
- TEST_FRACTION：要用于测试集的训练数据的比例。
如果指定了任一比例，则必须指定所有比例。这些比例之和必须等于 1.0。比例的默认值会因数据类型而异。了解详情。
```
"fractionSplit": {
  "trainingFraction": TRAINING_FRACTION,
  "validationFraction": VALIDATION_FRACTION,
  "testFraction": TEST_FRACTION
},
```
FilterSplit：

TRAINING_FILTER：与此过滤条件匹配的数据项用于训练集。
VALIDATION_FILTER：与此过滤条件匹配的数据项用于验证集。对于视频数据，该值必须为“-”。
TEST_FILTER：与此过滤条件匹配的数据项用于测试集。

这些过滤条件可以与 ml_use 标签或应用于数据的任何标签结合使用。详细了解如何使用 ml-use 标签和其他标签来过滤数据。

以下示例展示了如何将 filterSplit 对象与 ml_use 标签结合使用，其中包含验证集：

"filterSplit": {
"trainingFilter": "labels.aiplatform.googleapis.com/ml_use=training",
"validationFilter": "labels.aiplatform.googleapis.com/ml_use=validation",
"testFilter": "labels.aiplatform.googleapis.com/ml_use=test"
}

创建数据集

评估模型