MCMCSamples cannot read nested sampling output files correctly.

[xichenamoy]

Happened when execute this line:

sample = MCMCSamples(root=fileName)

Then get this error message:

ValueError: DataFrame constructor not properly called!

One thing worth to note is that 1.1.0 works well with the provided lensing example. So the error only happens when I change the 'root' equals to my own file name in the chains folder.

I also tried the standard PolyChord output file using the default gaussian example, it would report:

ValueError: Wrong number of items passed 5, placement implies 1

Previously when I used it with anesthetic 1.03, it can read the file ok. Now I upgraded it to version 1.1.0, run the same code it would report this. The output files are the standard MultiNest/PolyChord output files.

[williamjameshandley]

Hi @xichenamoy . Many thanks for spotting this.

Can you confirm that this only occurs if you try to use read nested sampling chains with MCMCSamples, and that reading them with NestedSamples does not have this issue?

@lukashergt, I think this was introduced in #14 when we restructured in terms of SampleReader. It might be worth just raising a more informative ValueError if one tries to read ns chains with MCMCSamples

[xichenamoy]

@williamjameshandley
Yes, I forgot to mention this. It works well with NestedSamples(), only occurs in MCMCSamples().

[lukashergt]

I think this was introduced in #14 when we restructured in terms of SampleReader

Probably, I never checked for the option of reading nested sampling files with MCMCSamples. Is that what we want?

[williamjameshandley]

I'm not sure. In theory NS files should be readable as MCMCSamples, but I don't know why you'd want to/if we should encourage it. Given that NestedSamples is a subclass of MCMCSamples you would get no benefit from loading them as MCMCSamples, so at the very least if we want to allow people to do this we should throw a warning. For now, the easiest solution would be to not support reading MN/PC files with MCMCSamples, but to throw a more informative error.

[lukashergt]

Fixed in #64