Open yannikschaelte opened 4 years ago
Whats the argument for storing configuration in hdf5? YAML would be so much more human readable/accessible and I don't think there are too many complex data formats.
the point would be to store in the file where we have the data also information on what was done -- i.e. what problem is it about, what optimizer was used, like done in parpe I think. yaml for human-readable output of course also a good idea. Will use dict anyway, so output to either is straightforward.
I guess that makes sense for the results file. However I don't think that makes sense when specifying the problem. I do see settings where it makes sense to reuse solver settings across problems but not data, or is this unrelated to #117?
Oh no, this thread here is not about specifying the problem, which should be done via yaml, completely agreed, but just about storing some additional meta information as output. #117 in contrast is about the input, defining what to use in the first place.
For storing them as attributes in hdf5, at least problem and optimizer should have functions e.g.
get_config -> dict[..., str/int/float/ndarray]
which return meta informtion, e.g. model name, optimizer settings, ...