Having a per file config dict

[matham]

There's something erking me here. Because you have a folder with logged files and a single config dict. The dict applies to all files in the folder? So for all the files we extract the same properties? Would we want to maybe have a per file config dict?

You're using plot_tensorboard only in generic pipeline rn so I understand that all files would be processed the same way. But, as a stand alone function, maybe it should instead accept a h5 filename, a list of attrs and settings and have the pipeline do the folder crawling etc. Partially it's also because HDFData accepts a buffer, so why not re-use it if we're reading multiple attrs from the same file. Or, why have it accept a buffer? If the same attr can be specified multiple times, then TensorboardWriter should be able to deal with that making it unique. If it can't, like it currently seems to be the case, the not just do it all in one HDFData?

But anyway that can all be left for later.

Originally posted by @matham in #1 (comment)