Reading a tensor from file in python which was saved using C++

[dibya-pati]

I'm having trouble reading/interpreting a tensor from a file using read_file() operation in Python. The tensor was saved using Save() ops in tensorflow CC API. I can read the underlying byte encoded tensor, but find no methods to convert that back to float(original content of the tensor). I have tried decode_raw() in python but it complaints about of not being of proper multiple of 4.
The above tensor comprises of a 3 channel nd array with float contents. Would appreciate any help in deciphering the content
More precisely would really appreciate on some details on how Save() ops in tensorflow CC API works?

[muddham]

@dibya-pati This question is better asked on StackOverflow since it is not a bug or feature request. There is also a larger community that reads questions there.If you think we've misinterpreted a bug, please comment again with a clear explanation, as well as all of the information requested in the issue template. Thanks!

[dibya-pati]

@muddham Thanks for the reply. I believe its more to do on documentation of an API . Its not a bug,But there are no documentation on how the tensor is encoded before writing to the disk , hence there is no way to understand how to decode it from the file using a different method other than Restore() in CC API. Thanks for the suggestion, I would definitely put a stackoverflow question But I would really appreciate if you could point out to me if there are any existing documentation of CC Save() API

[jvishnuvardhan]

@dibya-pati Could you provide a simple code to reproduce your issue? Thanks!

[dibya-pati]

Hi, Its not an issue, but documentation request i.e. I can't figure out from the documentation how the tensors are encoded before getting written to disk using Save() in TF CC API, So, I don't know how to read the written tensor file in python. Sorry, I couldn't get some time write a snippet to demonstrate this. But , if someone just could point to some example or any material that shows how its encoded, that will be it. I'll try to put a code snippet to better explain the problem

[Saduf2019]

@dibya-pati please confirm if the issue still persist

[MarkDaoust]

Hi @dibya-pati,

Personally, if I had to get a tensor-value from a cpp program to a python program I'd try:
serialize-tensor and parse-tensor first, since these have clear python implementations: serialize_tensor, and parse_tensor.

I'm no expert on the internals of tf-cpp or checkpoints, but I think ops::Save, and ops:SaveV2 are part of the checkpoint system.

[dibya-pati]

I'm still waiting for some API documentation on Save() ops in CC Tf. If you have some documentation please point me to it.

[rmothukuru]

@dibya-pati,
Sorry for the delayed response. Can you please refer to the documentation of Save and SaveV2 and let us know if this is what your are looking for? Thanks!

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler]

Closing as stale. Please reopen if you'd like to work on this further.