Store evidence

[vetlewi]

I've added an attribute to the ResultsNormalized to store the global evidence from the MultiNest results. Could be useful down the line.

[fzeiser]

Good idea.

How does this act when one employs the "stagewise" normalization? normalizeStagewise in EnsembleNormalizer? I havn't thought "enough" about it, but I assume the evidence should be multiplied somehow by the distance of the chosen B (-> Gg) from the experimental Gg. So this process here should modify the evidence I guess:

ompy/ompy/ensembleNormalizer.py

Lines 189 to 194 in 32572d5

	# sample B from the gaussian uncertainty for each nld
	B = self.normalizer_gsf.res.pars["B"]
	N = len(self.normalizer_gsf.res.samples["A"])
	self.normalizer_gsf.res.samples["B"] = scipynorm.rvs(loc=B[0],
	scale=B[1],
	size=N)

[vetlewi]

Good idea.

How does this act when one employs the "stagewise" normalization? normalizeStagewise in EnsembleNormalizer? I havn't thought "enough" about it, but I assume the evidence should be multiplied somehow by the distance of the chosen B (-> Gg) from the experimental Gg. So this process here should modify the evidence I guess:

ompy/ompy/ensembleNormalizer.py

Lines 189 to 194 in 32572d5

	# sample B from the gaussian uncertainty for each nld
	B = self.normalizer_gsf.res.pars["B"]
	N = len(self.normalizer_gsf.res.samples["A"])
	self.normalizer_gsf.res.samples["B"] = scipynorm.rvs(loc=B[0],
	scale=B[1],
	size=N)

Huh, haven't though about that. I guess it depends on the likelihood and prior to be separable (with regards to Gg)? I don't feel like I know enough Bayesian statistics to confidently give an answer :p

[fzeiser]

Unless this is very urgent, I'll think more about it after I've handed in my thesis in December. Writing on the main body of it now ;).

[vetlewi]

Unless this is very urgent, I'll think more about it after I've handed in my thesis in December. Writing on the main body of it now ;).

Not very urgent. Can always pull to a local branch for now.

[fzeiser]

I have difficulties to see how to correctly implement this for the sequential normalization. It somehow bothers me that we would combine a Bayesian model (nld normalization) with the frequentist Gg normalization and still call it a global evidence etc.

Suggestion: Simply store evidence only for NormalizerSimultan. (For the other normalization mode I somehow doubt that it makes sens.)

• remove evidence from NormalizerNLD.