bayesplain.priors.CorrelationPrior¶
-
class bayesplain.priors.CorrelationPrior(kappa, name=
'custom', rationale='')[source]¶ A stretched-beta prior on a correlation coefficient.
Parameterised by a width
kappa: the prior density is proportional to(1 - rho**2) ** (1/kappa - 1)on (-1, 1). Atkappa = 1it is flat, meaning every correlation from -1 to 1 is equally plausible before seeing data. Below 1 it concentrates near zero; above 1 it pushes mass toward the extremes.- Variables:¶
Methods
dist()Return the prior as a frozen scipy distribution on (-1, 1).