This is the new Amazon CloudFormation Template Reference Guide. Please update your bookmarks and links. For help getting started with CloudFormation, see the Amazon CloudFormation User Guide.
AWS::SageMaker::Algorithm TrainingSpecification
Defines how the algorithm is used for a training job.
Syntax
To declare this entity in your Amazon CloudFormation template, use the following syntax:
JSON
{ "MetricDefinitions" :[ MetricDefinition, ... ], "SupportedHyperParameters" :[ HyperParameterSpecification, ... ], "SupportedTrainingInstanceTypes" :[ String, ... ], "SupportedTuningJobObjectiveMetrics" :[ HyperParameterTuningJobObjective, ... ], "SupportsDistributedTraining" :Boolean, "TrainingChannels" :[ ChannelSpecification, ... ], "TrainingImage" :String, "TrainingImageDigest" :String}
YAML
MetricDefinitions:- MetricDefinitionSupportedHyperParameters:- HyperParameterSpecificationSupportedTrainingInstanceTypes:- StringSupportedTuningJobObjectiveMetrics:- HyperParameterTuningJobObjectiveSupportsDistributedTraining:BooleanTrainingChannels:- ChannelSpecificationTrainingImage:StringTrainingImageDigest:String
Properties
MetricDefinitions-
A list of
MetricDefinitionobjects, which are used for parsing metrics generated by the algorithm.Required: No
Type: Array of MetricDefinition
Maximum:
40Update requires: Replacement
SupportedHyperParameters-
A list of the
HyperParameterSpecificationobjects, that define the supported hyperparameters. This is required if the algorithm supports automatic model tuning.>Required: No
Type: Array of HyperParameterSpecification
Maximum:
100Update requires: Replacement
SupportedTrainingInstanceTypes-
A list of the instance types that this algorithm can use for training.
Required: Yes
Type: Array of String
Update requires: Replacement
SupportedTuningJobObjectiveMetrics-
A list of the metrics that the algorithm emits that can be used as the objective metric in a hyperparameter tuning job.
Required: No
Type: Array of HyperParameterTuningJobObjective
Update requires: Replacement
SupportsDistributedTraining-
Indicates whether the algorithm supports distributed training. If set to false, buyers can't request more than one instance during training.
Required: No
Type: Boolean
Update requires: Replacement
TrainingChannels-
A list of
ChannelSpecificationobjects, which specify the input sources to be used by the algorithm.Required: Yes
Type: Array of ChannelSpecification
Minimum:
1Maximum:
8Update requires: Replacement
TrainingImage-
The Amazon ECR registry path of the Docker image that contains the training algorithm.
Required: Yes
Type: String
Maximum:
255Update requires: Replacement
TrainingImageDigest-
An MD5 hash of the training algorithm that identifies the Docker image used for training.
Required: No
Type: String
Pattern:
^[Ss][Hh][Aa]256:[0-9a-fA-F]{64}$Maximum:
72Update requires: Replacement