Can't serialise objects with numpy elements

[brianbienvenu]

Trying to write out an object with a numpy element (ndarray) throws this:

  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 428, in dump
    CBOREncoder(fp, **kwargs).encode(obj)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 386, in encode
    encoder(self, obj)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 49, in wrapper
    func(encoder, value, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 107, in encode_array
    encoder.encode(item)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 386, in encode
    encoder(self, obj)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 49, in wrapper
    func(encoder, value, *args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 115, in encode_map
    encoder.encode(val)
  File "/usr/local/lib/python3.7/dist-packages/cbor2/encoder.py", line 384, in encode
    raise CBOREncodeError('cannot serialize type %s' % obj_type.__name__)
cbor2.encoder.CBOREncodeError: cannot serialize type ndarray

Can this be handled better?

[Sekenre]

Hi @brianbienvenu have a look at https://cbor2.readthedocs.io/en/latest/customizing.html#using-the-cbor-tags-for-custom-types and https://docs.scipy.org/doc/numpy/reference/generated/numpy.ndarray.dumps.html#numpy.ndarray.dumps

Unfortunately ndarrays are wrappers around multi-dimensional buffers of C-types so it might be easiest for you to pickle them and save the resulting bytes as a single tagged item than encode them in CBOR.

Alternatively you could serialize them item-by-item as lists of lists of integers. This will be slow on both encode and decode. ndarrays also come with a bunch of metadata that needs to be saved and recreated.