How to pickle an Ptxt?

[Shakemilktea]

Hello, I got some error when pickling ptxt. This is the error:

Traceback (most recent call last):
  File "E:\NCKU\ICLab\code\Pyfhel-master\examples\Demo_4_SaveNRestore.py", line 172, in <module>
    pkls_ptxt = pickle.dumps(p)
  File "<stringsource>", line 2, in Pyfhel.PyPtxt.PyPtxt.__reduce_cython__
TypeError: no default __reduce__ due to non-trivial __cinit__

I'd like to encode the deep learning model's weights, then I can perform operations between the encrypted image and the encoded weights. However, when I attempt to encode the weights, a MemoryError might occur (I guess, because the code will stop without raising an error.) Therefore, I intend to pickle the encoded weights (Ptxt) to avoid overloading the RAM. Pickling ctxt doing well in #81, but pickling ptxt gets error.

Do you have any advice?

[ibarrond]

@Shakemilktea The encoding operation is light-weight compared to the homomorphic operations. That being said, you could:

• Pickle the weights as numpy objects.

• Forl Pyfhel, implement the necessary pickling mechanism for PyPtxt and merge it on the main repo. This seems doable:

  1. Create a __reduce__ method in PyPtxt.pyx that does exactly what PyCtxt.__reduce__ does
  2. Test it and add some example of its use in an examples file.

I'd be happy to provide further assistance if you decide to implement it.

[Shakemilktea]

Thanks for your advice! It looks good when multiplying Numpy and Ctxt!
Regarding the pickling of Ptxt, I haven't had experience with Cython before, so I need to study it first.