The to_input_req method of the TuningJobCompletionCriteriaConfig doesn't work and makes it unusable

[konradsemsch]

Describe the bug

When trying to specify the TuningJobCompletionCriteriaConfig and using the to_input_req method on the instantiated class object (which the HPO tuner does when the fit() function is called), it throws an error, which prevents from using it. This is because in the to_input_req values are assigned to non-existent keys of a dictionary.

Additionally, the TuningJobCompletionCriteriaConfig is not documentation in the Python SDK: https://sagemaker.readthedocs.io/en/stable/api/training/tuner.html

To reproduce
A clear, step-by-step set of instructions to reproduce the bug.

from sagemaker.tuner import (
    TuningJobCompletionCriteriaConfig, 
    )
test = TuningJobCompletionCriteriaConfig(
        max_number_of_training_jobs_not_improving=10,
        complete_on_convergence=True,
        target_objective_metric_value=0,
        )
test.to_input_req()

Results in:

When we try to do it only for the target_objective_metric_value it works as expected. This is because there is just one level of the dictionary for this argument in the completion_criteria_config in the to_input_req which confirms my assumption in the description of the issue.

Expected behavior

Given this input:

from sagemaker.tuner import (
    TuningJobCompletionCriteriaConfig, 
    )
test = TuningJobCompletionCriteriaConfig(
        max_number_of_training_jobs_not_improving=10,
        complete_on_convergence=True,
        target_objective_metric_value=0,
        )
test.to_input_req()

I would expect the function to return:

{'BestObjectiveNotImproving': {'MaxNumberOfTrainingJobsNotImproving': 10},
 'TargetObjectiveMetricValue': 0,
 'ConvergenceDetected': {'CompleteOnConvergence': 'Enabled'}}

This can be achieved by modifying the to_input_req of the TuningJobCompletionCriteriaConfig class to:

    def to_input_req(self):
        """Converts the ``self`` instance to the desired input request format.

        Examples:
            >>> completion_criteria_config = TuningJobCompletionCriteriaConfig(
                max_number_of_training_jobs_not_improving=5
                complete_on_convergence = True,
                target_objective_metric_value = 0.42
            )
            >>> completion_criteria_config.to_input_req()
            {
                "BestObjectiveNotImproving": {
                    "MaxNumberOfTrainingJobsNotImproving":5
                },
                "ConvergenceDetected": {
                    "CompleteOnConvergence": "Enabled",
                },
                "TargetObjectiveMetricValue": 0.42
            }

        Returns:
            dict: Containing the completion criteria configurations.
        """

        completion_criteria_config = {}
        if self.max_number_of_training_jobs_not_improving is not None:
            completion_criteria_config[BEST_OBJECTIVE_NOT_IMPROVING] = {}  ##### <---- CHANGED
            completion_criteria_config[BEST_OBJECTIVE_NOT_IMPROVING][
                MAX_NUMBER_OF_TRAINING_JOBS_NOT_IMPROVING
            ] = self.max_number_of_training_jobs_not_improving

        if self.target_objective_metric_value is not None:
            completion_criteria_config[
                TARGET_OBJECTIVE_METRIC_VALUE
            ] = self.target_objective_metric_value

        if self.complete_on_convergence is not None:
            completion_criteria_config[CONVERGENCE_DETECTED] = {}  ##### <---- CHANGED
            completion_criteria_config[CONVERGENCE_DETECTED][COMPLETE_ON_CONVERGENCE_DETECTED] = (
                "Enabled" if self.complete_on_convergence else "Disabled"
            )

        return completion_criteria_config

System information
A description of your system. Please provide:

• SageMaker Python SDK version: '2.140.1'
• Python version: 3.10.9
• CPU or GPU: CPU
• Custom Docker image (Y/N): N

[martinRenou]

This can probably be closed now that #3744 is merged?

[martinRenou]

Closing as fixed. Thanks!