JohannesBuchner / PyMultiNest

Pythonic Bayesian inference and visualization for the MultiNest Nested Sampling Algorithm and PyCuba's cubature algorithms.
http://johannesbuchner.github.io/PyMultiNest/
Other
191 stars 87 forks source link

local ln(Z) vs. strictly local ln(Z) #207

Closed rpoleski closed 2 years ago

rpoleski commented 2 years ago

With multimode function turned on, I'm getting the same values of Strictly Local Log-Evidence and Local Log-Evidence, but their errors are different. Does anybody know what is the difference between them?

JohannesBuchner commented 2 years ago

May have to do with Ztol https://github.com/JohannesBuchner/MultiNest/blob/master/README

rpoleski commented 2 years ago

I haven't found anything directly related to Ztol. I've made some tests with toy models with known probability of each mode. I've calculated probabilities as p_i = Z_i/np.sum(Z_i) and its error: sigma_p_i = p_i * sigma_ln_Z_i (i.e., ignoring uncertainty of np.sum(Z_i)). The uncertainties of Local Log-Evidence are closer to the scatter of p_i around known values than for its strict version.

JohannesBuchner commented 2 years ago

I am closing this as it is not a PyMultiNest issue.