copy of input_receiver.features

[wu-yy]

deepcopy of input_receiver.features avoid changing input_receiver.features in model_fn

[k-w-w]

Thanks for the PR!

I think this counts as a shallow copy, since the values aren't being copied before being added to the new dict. You could replace the loop and use: copy.copy(input_receiver.features)

Also, is there an issue associated with this change? Typically, the input_receiver_fn produces a new features dict that isn't used anywhere else, which is why modifications to it would be generally acceptable. (but for code cleanliness, I agree that making a copy before passing it to the model function is better).

[wu-yy]

@k-w-w thanks for your reply， We encountered this problem because the incoming features were modified in model_fn， so, the placeholder which in savdemodel is not matched the input_receiver_fn. I will change the PR and use copy.copy(input_receiver.features).

[gbaned]

Hi @k-w-w Can you please review this PR ? Thank you!

[mihaimaruseac]

No more updates to Estimator, except to ensure compat with TF releases, when needed