How is the logl value computed wrt to the Sherpa Wstat ?

[facero]

• BXA version: 4.1.1
• UltraNest version: 3.5.7
• Python version: 3.9
• Xspec or Sherpa and version: ciao 4.15
• Operating System: MacOS 12.6

Description

As a way to understand how the NS works and to visualize the evolution of likelihood as a function of iterations, I'm looking at the logL column in the weighted_post.txt.

I'm using unbinned X-ray spectral data and the set_stat("wstat") in Sherpa. The background is extracted from the same observation. The BXA file weighted_post.txt looks this :

In this case, the BXA maximized the logL value (the last rows with higher value are better) whereas in the sherpa or Xspec implementation, we minize Cstat or Wstat. How is this logL value is calculated ? Can it be related to Wstat value computed by sherpa ?

One the thing I'm trying is comparing the Wstat values from sherpa's fit to the solutions found by BXA.

[JohannesBuchner]

Have a look at the log-likelihood definition:

with pyxspec: https://github.com/JohannesBuchner/BXA/blob/master/bxa/xspec/solver.py#L160

with sherpa: https://github.com/JohannesBuchner/BXA/blob/master/bxa/sherpa/solver.py#L123

[facero]

Ah ok so it keeps the original Sherpa choice and just mulitply by -0.5. Thanks