`HyperparameterTuner.attach()` does not get `use_spot_instances` setting

[dz902]

Describe the bug

When using the following:

PARENT_TUNER = HyperparameterTuner.attach(
    tuning_job_name = PARENT_TUNING_JOB_NAME
)

...on a tuning job where its job definition has:

...
        "StoppingCondition": {
            "MaxRuntimeInSeconds": 3600,
            "MaxWaitTimeInSeconds": 7200
        },
        "EnableNetworkIsolation": false,
        "EnableInterContainerTrafficEncryption": false,
        "EnableManagedSpotTraining": true
...

The max_wait and use_spot_instances setting are both None. I traced back to:

sagemaker-python-sdk/src/sagemaker/estimator.py

Lines 811 to 874 in 481719f

	def _prepare_init_params_from_job_description(cls, job_details, model_channel_name=None):
	"""Convert the job description to init params that can be handled by the
	class constructor
	Args:
	job_details: the returned job details from a describe_training_job
	API call.
	model_channel_name (str): Name of the channel where pre-trained
	model data will be downloaded.
	Returns:
	dictionary: The transformed init_params
	"""
	init_params = dict()
	init_params["role"] = job_details["RoleArn"]
	init_params["instance_count"] = job_details["ResourceConfig"]["InstanceCount"]
	init_params["instance_type"] = job_details["ResourceConfig"]["InstanceType"]
	init_params["volume_size"] = job_details["ResourceConfig"]["VolumeSizeInGB"]
	init_params["max_run"] = job_details["StoppingCondition"]["MaxRuntimeInSeconds"]
	init_params["input_mode"] = job_details["AlgorithmSpecification"]["TrainingInputMode"]
	init_params["base_job_name"] = base_from_name(job_details["TrainingJobName"])
	init_params["output_path"] = job_details["OutputDataConfig"]["S3OutputPath"]
	init_params["output_kms_key"] = job_details["OutputDataConfig"]["KmsKeyId"]
	if "EnableNetworkIsolation" in job_details:
	init_params["enable_network_isolation"] = job_details["EnableNetworkIsolation"]
	has_hps = "HyperParameters" in job_details
	init_params["hyperparameters"] = job_details["HyperParameters"] if has_hps else {}
	if "AlgorithmName" in job_details["AlgorithmSpecification"]:
	init_params["algorithm_arn"] = job_details["AlgorithmSpecification"]["AlgorithmName"]
	elif "TrainingImage" in job_details["AlgorithmSpecification"]:
	init_params["image_uri"] = job_details["AlgorithmSpecification"]["TrainingImage"]
	else:
	raise RuntimeError(
	"Invalid AlgorithmSpecification. Either TrainingImage or "
	"AlgorithmName is expected. None was found."
	)
	if "MetricDefinitons" in job_details["AlgorithmSpecification"]:
	init_params["metric_definitions"] = job_details["AlgorithmSpecification"][
	"MetricsDefinition"
	]
	if "EnableInterContainerTrafficEncryption" in job_details:
	init_params["encrypt_inter_container_traffic"] = job_details[
	"EnableInterContainerTrafficEncryption"
	]
	subnets, security_group_ids = vpc_utils.from_dict(job_details.get(vpc_utils.VPC_CONFIG_KEY))
	if subnets:
	init_params["subnets"] = subnets
	if security_group_ids:
	init_params["security_group_ids"] = security_group_ids
	if "InputDataConfig" in job_details and model_channel_name:
	for channel in job_details["InputDataConfig"]:
	if channel["ChannelName"] == model_channel_name:
	init_params["model_channel_name"] = model_channel_name
	init_params["model_uri"] = channel["DataSource"]["S3DataSource"]["S3Uri"]
	break
	return init_params

It seems use_spot_instances and max_wait do not get carried over to newly created estimator.

To reproduce

See above.

Expected behavior

use_spot_instances and max_wait etc. should all be carried over to newly attach()ed tuner. This also affects warm start helpers like identical_data_and_algorithm().

If applicable, add screenshots or logs to help explain your problem.

**System information**
A description of your system. Please provide:
- **SageMaker Python SDK version**: v2.0.0
- **Framework name (eg. PyTorch) or algorithm (eg. KMeans)**:
- **Framework version**:
- **Python version**:
- **CPU or GPU**:
- **Custom Docker image (Y/N)**: N, official image classification image

**Additional context**
Add any other context about the problem here.

[dz902]

Also inputs from parent tuner is ignored in newly created tuner when calling fit() without the inputs arg. In official docs fit() was called with input as an empty set input={} or just plain fit(), which to me indicates that there is a way to inherit of set inputs, but it doesn't seem to be the case.

[dz902]

Until this is fixed, a workaround:

PARENT_TUNER.estimator.max_wait = 7200 # !!!
PARENT_TUNER.estimator.use_spot_instances = True # !!!

[icywang86rui]

The fix is merged and released in v2.5.5. @dz902 thanks for reporting the bug!