"Exception: Object of type ParameterString is not JSON serializable" when passing ParameterString as part of input in hyperparametertuner fit method.

Describe the bug When running the HyperparameterTuner.fit method, with ParameterString as part of input, the method errors out with following: Exception: Object of type ParameterString is not JSON serializable. Even replacing the ParameterStrings with, strings produces the same error.

To reproduce bucket = "ml-bucket" prefix = "sample-prefix/"

image_uri = sagemaker.image_uris.retrieve( framework="xgboost", region=region, version="1.0-1", py_version="py3", instance_type=training_instance_type, ) role = sagemaker.session.get_execution_role(sagemaker_session) train_instance_count=1 training_instance_type="ml.m4.4xlarge" estimator_prefix="generic-estimator-prefix" estimator_output_path=f"s3://{bucket}/{estimator_prefix}" train_data_path = f's3://{bucket}/{prefix}train' test_data_path = f's3://{bucket}/{prefix}test'

train_data_path = ParameterString( name="TrainDataPath", default_value=train_data_path, ) test_data_path = ParameterString( name="TestDataPath", default_value=test_data_path, )

xgb = sagemaker.estimator.Estimator( image_uri=image_uri, role=role, instance_count=train_instance_count, instance_type=training_instance_type, output_path=estimator_output_path, base_job_name="generic-base-job-name", hyperparameters=HPs, debugger_hook_config=False, # disable SageMaker Debugger disable_profiler=True # Specifies whether Debugger monitoring and profiling will be disabled )

search_grid = { "alpha": ContinuousParameter(0.01, 10, scaling_type="Logarithmic"), "eta": ContinuousParameter(0.1, 0.5), "gamma": ContinuousParameter(0, 5), "min_child_weight": ContinuousParameter(0, 120), "max_depth": IntegerParameter(1, 7), "subsample": ContinuousParameter(0.5, 1) }

tuner = HyperparameterTuner( xgb, objective_metric_name="validation:auc", objective_type="Maximize", hyperparameter_ranges=search_grid, max_jobs=2, max_parallel_jobs=2, early_stopping_type="Auto", strategy="Bayesian" )

step_args = tuner.fit(inputs={ "train": TrainingInput( s3_data=train_data_path, ), "validation": TrainingInput( s3_data=test_data_path ), }, )

Even trying above without setting train_data_path, and test_data_path as ParameterStrings fails with same error.

Expected behavior Tuner successfully runs tuning job. Screenshots or logs tuner_job_fail

System information AWS Codebuild

SageMaker Python SDK version: 2.104.0
Framework name (eg. PyTorch) or algorithm (eg. KMeans): Bayesian HyperParameterTuner, XGBoost Sagemaker Model
Python version: 3.8

aws / sagemaker-python-sdk

"Exception: Object of type ParameterString is not JSON serializable" when passing ParameterString as part of input in hyperparametertuner fit method. #3320