Hi,
Thank you for your inputs!
Can I ask: Why is pm.Uniform discouraged?
The data I am working with are count data from a nuclear experiment. The code I have presented here is a correction code for the detector response (called resmat in the code). The signal hitting the detector gets transformed by the detector response and then the output signal is different than the original signal. So this code is correcting for this. The response resmat is a matrix and the problem is: reco = truth@resmat. The response matrix is not invertible so there is many possible solutions and we are only interested in the positive solutions, since negative counts are ‘unphysical’ when looking at count data. That is why I use a uniform distribution, to set a lower bound to zero for all parameters. A normal distribution allows for the negative solutions, and for some of the parameters the negative solutions are the most probable solution.
But I am very open to suggestions if there is any other priors that can be used in this case.