Rekognition / Client / create_project_version

create_project_version#

Rekognition.Client.create_project_version(**kwargs)#

Creates a new version of a model and begins training. Models are managed as part of an Amazon Rekognition Custom Labels project. The response from CreateProjectVersion is an Amazon Resource Name (ARN) for the version of the model.

Training uses the training and test datasets associated with the project. For more information, see Creating training and test dataset in the Amazon Rekognition Custom Labels Developer Guide.

Note

You can train a model in a project that doesn’t have associated datasets by specifying manifest files in the TrainingData and TestingData fields.

If you open the console after training a model with manifest files, Amazon Rekognition Custom Labels creates the datasets for you using the most recent manifest files. You can no longer train a model version for the project by specifying manifest files.

Instead of training with a project without associated datasets, we recommend that you use the manifest files to create training and test datasets for the project.

Training takes a while to complete. You can get the current status by calling DescribeProjectVersions. Training completed successfully if the value of the Status field is TRAINING_COMPLETED.

If training fails, see Debugging a failed model training in the Amazon Rekognition Custom Labels developer guide.

Once training has successfully completed, call DescribeProjectVersions to get the training results and evaluate the model. For more information, see Improving a trained Amazon Rekognition Custom Labels model in the Amazon Rekognition Custom Labels developers guide.

After evaluating the model, you start the model by calling StartProjectVersion.

This operation requires permissions to perform the rekognition:CreateProjectVersion action.

Request Syntax

response = client.create_project_version(
    ProjectArn='string',
    VersionName='string',
    OutputConfig={
        'S3Bucket': 'string',
        'S3KeyPrefix': 'string'
    },
    TrainingData={
        'Assets': [
            {
                'GroundTruthManifest': {
                    'S3Object': {
                        'Bucket': 'string',
                        'Name': 'string',
                        'Version': 'string'
                    }
                }
            },
        ]
    },
    TestingData={
        'Assets': [
            {
                'GroundTruthManifest': {
                    'S3Object': {
                        'Bucket': 'string',
                        'Name': 'string',
                        'Version': 'string'
                    }
                }
            },
        ],
        'AutoCreate': True|False
    },
    Tags={
        'string': 'string'
    },
    KmsKeyId='string'
)

Parameters:

ProjectArn (string) –
[REQUIRED]

The ARN of the Amazon Rekognition Custom Labels project that manages the model that you want to train.
VersionName (string) –
[REQUIRED]

A name for the version of the model. This value must be unique.
OutputConfig (dict) –
[REQUIRED]

The Amazon S3 bucket location to store the results of training. The S3 bucket can be in any AWS account as long as the caller has s3:PutObject permissions on the S3 bucket.
- S3Bucket (string) –
  
  The S3 bucket where training output is placed.
- S3KeyPrefix (string) –
  
  The prefix applied to the training output files.
TrainingData (dict) –
Specifies an external manifest that the services uses to train the model. If you specify TrainingData you must also specify TestingData. The project must not have any associated datasets.
- Assets (list) –
  
  A Sagemaker GroundTruth manifest file that contains the training images (assets).
  - (dict) –
    
    Assets are the images that you use to train and evaluate a model version. Assets can also contain validation information that you use to debug a failed model training.
    - GroundTruthManifest (dict) –
      
      The S3 bucket that contains an Amazon Sagemaker Ground Truth format manifest file.
      - S3Object (dict) –
        
        Provides the S3 bucket name and object name.
        
        The region for the S3 bucket containing the S3 object must match the region you use for Amazon Rekognition operations.
        
        For Amazon Rekognition to process an S3 object, the user must have permission to access the S3 object. For more information, see How Amazon Rekognition works with IAM in the Amazon Rekognition Developer Guide.
        
        Bucket (string) –
        
        Name of the S3 bucket.
        
        Name (string) –
        
        S3 object key name.
        
        Version (string) –
        
        If the bucket is versioning enabled, you can specify the object version.
TestingData (dict) –
Specifies an external manifest that the service uses to test the model. If you specify TestingData you must also specify TrainingData. The project must not have any associated datasets.
- Assets (list) –
  
  The assets used for testing.
  - (dict) –
    
    Assets are the images that you use to train and evaluate a model version. Assets can also contain validation information that you use to debug a failed model training.
    - GroundTruthManifest (dict) –
      
      The S3 bucket that contains an Amazon Sagemaker Ground Truth format manifest file.
      - S3Object (dict) –
        
        Provides the S3 bucket name and object name.
        
        The region for the S3 bucket containing the S3 object must match the region you use for Amazon Rekognition operations.
        
        For Amazon Rekognition to process an S3 object, the user must have permission to access the S3 object. For more information, see How Amazon Rekognition works with IAM in the Amazon Rekognition Developer Guide.
        
        Bucket (string) –
        
        Name of the S3 bucket.
        
        Name (string) –
        
        S3 object key name.
        
        Version (string) –
        
        If the bucket is versioning enabled, you can specify the object version.
- AutoCreate (boolean) –
  
  If specified, Amazon Rekognition Custom Labels temporarily splits the training dataset (80%) to create a test dataset (20%) for the training job. After training completes, the test dataset is not stored and the training dataset reverts to its previous size.
Tags (dict) –
A set of tags (key-value pairs) that you want to attach to the model.
- (string) –
  - (string) –
KmsKeyId (string) –
The identifier for your AWS Key Management Service key (AWS KMS key). You can supply the Amazon Resource Name (ARN) of your KMS key, the ID of your KMS key, an alias for your KMS key, or an alias ARN. The key is used to encrypt training and test images copied into the service for model training. Your source images are unaffected. The key is also used to encrypt training results and manifest files written to the output Amazon S3 bucket ( OutputConfig).

If you choose to use your own KMS key, you need the following permissions on the KMS key.
- kms:CreateGrant
- kms:DescribeKey
- kms:GenerateDataKey
- kms:Decrypt
If you don’t specify a value for KmsKeyId, images copied into the service are encrypted using a key that AWS owns and manages.

Return type:

dict

Returns:

Response Syntax

{
    'ProjectVersionArn': 'string'
}

Response Structure

(dict) –
- ProjectVersionArn (string) –
  
  The ARN of the model version that was created. Use DescribeProjectVersion to get the current status of the training operation.

Exceptions

Rekognition.Client.exceptions.ResourceInUseException
Rekognition.Client.exceptions.ResourceNotFoundException
Rekognition.Client.exceptions.LimitExceededException
Rekognition.Client.exceptions.InvalidParameterException
Rekognition.Client.exceptions.AccessDeniedException
Rekognition.Client.exceptions.InternalServerError
Rekognition.Client.exceptions.ThrottlingException
Rekognition.Client.exceptions.ProvisionedThroughputExceededException
Rekognition.Client.exceptions.ServiceQuotaExceededException