pickling / unpickling of data object about madmom

[fdlm]

1) would be nice, but we need to consider what to save and what not.
2) is also good. Since all the data classes are also numpy arrays, you can save the data using np.save if you want.
3) is the worst solution - keeping track of what is possible and what is not while using the library is cumbersome and weird.

I think if we are to implement pickling it should work seemlessly - meaning that if I pickle and unpickle the data object, it is exactly the same. This would require a lot of information (stft, frames, signal...) to be stored, as you mentioned. But if you want to save a stripped-down version of the data, you can always use np.save.

I would choose 2) for now, we can add pickling later anyways without API break.

[superbock]

Ok, I chose option 2), since it is the safest right now. See PR #81.