How to write a sagemaker tensorflow input_handler() that returns a numpy array?

Question

Posting it intentionally with a similar caption as this question: How to correctly write a sagemaker tensorflow input_handler() that returns a numpy array? because the sugge