File I/O

[OliverAh]

The data-structures introduced and used by the dwave-ocean-sdk mostly consist of several substructures that might or might not be native to python, e.g. the class dwave.system.EmbeddingComposite [1] consists of properties, such as EmbeddingComposite.child and EmbeddingComposite.parameters, and methods, such as EmbeddingComposite.sample_qubo(...), or the class dimod.SampleSet [2] consists of properties, such as SampleSet.record, and methods, such as SampleSet.lowest(...).
However, it is currently not possible to simply write to files, and read from files in a different script, instaces of them. Please do not consider this a limitation to my examples. I know there are a lot more.

In my opinion a missing feature of the dwave-ocean-sdk is write-to and read-from file(s) all kinds of data-structures related to the dwave packages. For example NumPy offers, apart from many other functions, the functions writetxt(filename, array) and loadtxt(filename) [3] to easily write arrays to files and also read them in from files.

So far, I also tried to use the python package pickle [4] (functions pickle.dump(obj, file) for writing and pickle.load(file) for reading). On a simple test-case that worked very well for an instance of a dimod.Sampleset, but failed for an instance of a dwave.system.DWaveSampler, due to some subsub...subsubstructure within, which can not "be pickled".
One could of course also go through all data-structures and recursively all their substructures and decide what data is of interest, but this is of course ... cumbersome ...
I think the pickle package is offering even more than necessary. I think it would be sufficient to I/O data (properties in the examples above), while methods should be known by the reading script from the corresponding packages. But it should be possible to treat the read data-structures as the ones they were written, e.g. a written instance of a dwave.system.EmbeddingComposite should be readable by a different script that imported the packages of the dwave-ocean-sdk, and treat the read data again as an instance of a dwave.system.EmbeddingComposite.

Why do I think such a feature is missing:
Currently a lot of research is going on to improve applicability of quantum annealing to real-world problems. But since there are quite some subtleties, there also is a lot of trial and error. In this regard, I think it would be beneficial for the community to exchange all kinds of findings and especially data. I might even dare to say that the inputs are more interesting than the results, and of course the combination of both would be the most valuable.

1 - https://docs.ocean.dwavesys.com/en/stable/docs_system/reference/composites.html#embeddingcomposite
2 - https://docs.ocean.dwavesys.com/en/stable/docs_dimod/reference/sampleset.html#id1
3 - https://numpy.org/doc/stable/reference/routines.io.html
4 - https://docs.python.org/3/library/pickle.html#