AWS Sagemaker API reference#

This package contains things that are useful when extending Flytekit.


The algorithm name is used for deciding which pre-built image to point to.

AlgorithmSpecification([algorithm_name, ...])

Specifies the training algorithm to be used in the training job This object is mostly a pass-through, with a couple of exceptions include: (1) in Flyte, users don't need to specify TrainingImage; either use the built-in algorithm mode by using Flytekit's Simple Training Job and specifying an algorithm name and an algorithm version or (2) when users want to supply custom algorithms they should set algorithm_name field to CUSTOM.


param List[str] values

list of strings representing categorical values

ContinuousParameterRange(max_value, ...)

param float max_value


str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str


The distribution framework is used for determining which underlying distributed training mechanism to use.

DistributedTrainingContext(current_host, ...)

HPOJob(max_number_of_training_jobs, ...[, ...])

HPOJob Configuration should be used to configure the HPO Job.



The specification of the hyperparameter tuning process


HyperparameterTuningObjective is a data structure that contains the target metric and the objective of the hyperparameter tuning.




Specifies the type of content for input data.


When using FILE input mode, different SageMaker built-in algorithms require different file types of input data See

IntegerParameterRange(max_value, min_value, ...)

param int max_value


Initializes a new ParameterRangeOneOf. :param Union[IntegerParameterRange, ContinuousParameterRange, CategoricalParameterRange] param: One of the supported parameter ranges.

SagemakerBuiltinAlgorithmsTask(*args, **kwargs)

Implements an interface that allows execution of a SagemakerBuiltinAlgorithms.

SagemakerCustomTrainingTask(*args, **kwargs)

Allows a custom training algorithm to be executed on Amazon Sagemaker.

SagemakerHPOTask(*args, **kwargs)

param task_type

defines a unique task-type for every new extension. If a backend plugin is required then

SagemakerTrainingJobConfig(...[, ...])

Configuration for Running Training Jobs on Sagemaker.


TrainingJobResourceConfig(instance_type, ...)

TrainingJobResourceConfig is a pass-through, specifying the instance type to use for the training job, the number of instances to launch, and the size of the ML storage volume the user wants to provision Refer to SageMaker official doc for more details: