The role of reparametrized_params variable in various distributions

I would like to introduce a new distribution class for an AR(1) process following the existing implementation of the GaussianRandomWalk distribution. However, I do not understand the role of the ‘reparametrized_params’, e.g.

reparametrized_params = [‘scale’]

Could someone clarify this part a bit?

I noticed that other distributions contain similar lists but I do not understand which variables should be contained here and which not.

For example Gamma distribution has two positive definite variables but only ‘rate’ is listed as re-parameterized.

class Gamma(Distribution):
arg_constraints = {‘concentration’: constraints.positive,
‘rate’: constraints.positive}
support = constraints.positive
reparametrized_params = [‘rate’]

@pyroman It is not important, we only use it for testing. :smiley:

@fehiepsi Thanks, good to know. :sweat_smile: