Testing FBU on gamma ray spectra

[valamaria89]

Hi,

I am currently testing the FBU-library to unfold gamma ray spectra (taken from nuclear physics experiments).

I have been reading the article called Fully Bayesian Unfolding by Georgios Choudalakis, and to my understanding this library is built on this article. The article describes another sampling approach than I see used in the library?
In section 5.3 it says the following: One needs to be careful with MCMC sampling (Sec. 5.2), because it does not sample the T-space uniformly. This invalidates Eq. 22.

So my question is: When using the NUTS sampler (as is automatically initiated when using PyMC3), can that sample the T-space in an unwanted way?
As the code stands now, I don't see any problem using the NUTS sampler as each bin is sampled individually, so the trace consists of a posterior distribution for each bin and they are continuous. But I have to ask to be sure, because I do not really understand why the article says that the T-space has to be sampled uniformly...

[clementhelsens]

Dear @valamaria89 ,

yes, you are correct, the sampling of the bins is done individually here, thus the T-Space should not be sampled in an unwanted way.