Tensorflow estimator script mode not working if we pass HyperParameters as part of TrainingStep

[the-elancier]

Hello,

I was trying to use Tensorflow estimator in script mode and I was trying to pass the hyper-parameters as part of stepfunctions.steps.TrainingStep as follows:

But, the training job failed with the following error

Upon investigation I found out that other hyper-parameters that the estimator attaches on its own were not going (ex - 'sagemaker_program', 'sagemaker_submit_directory' etc.)

For example, the following image is the set of hyper-parameters for the job when we give it as part of TrainingStep:

But, if we give the hyper-parameters as part of the estimator:

The reason why I'd want to give hyper-parameters as part of the TrainingStep is that I'd be able to use stepfunctions.inputs.ExecutionInput in order to provide inputs dynamically that might be useful in the script.

[shivlaks]

fixed through #144 which enables the capability to specify hyperparameters in the TrainingStep constructor and merges them with hyperparameters established in the estimator.