-
Notifications
You must be signed in to change notification settings - Fork 2
Model behavior
Jonas Eschle edited this page Sep 2, 2019
·
2 revisions
A model can be normalized or not and has three different methods to access it's underlying function
If a model is normalized, the obs
Space will be taken automatically as normalization range
if no explicit Space
is given to the "normalization keyword (WIP).
Normalization means that the integral over the norm range is one. This is done column wise in case of a N-dim function.
There are three methods which fulfill similar needs, pdf
, value
and func
. (It is possible that in future versions one or more will be dropped). The table shows the return whether it is normalized ("norm") or not ("non-norm") depending on whether the model is normed ("yes"/"no") and on which of the three function is called.
normed | pdf |
value |
func |
---|---|---|---|
yes | norm | norm | non-norm |
no | error | non-norm | non-norm |
Rational for the functions
-
pdf
serves the purpose to guarantee a normalized model and does a simple check if the model is normed. E.g. for building a likelihood, a PDF is required. -
value
allows to be agnostic on the normalization and should be used whenever the normalization is not strictly required or "forbidden" -
func
is the unnormalized function. While in generalvalue
should be preferred, sometimes intentionally the unnormalized value of the model is required. It somewhat replaces theunnormalized_pdf
and should, in future, be used internally (?).