From the log p.d.f plot, BiSymmetricLogUniform seems to be a uniform distribution instead of log-uniform. Although samples from BiSymmetricLogUniform.samples() are distributed as BiSymmetricLogUniform, the NUTS samples are not, see plots below. The key to make BiSymmetricLogUniform working for NUTS is to correctly implement the log_prob method, and icdf for nested sampling of jaxns.
From the log p.d.f plot,
BiSymmetricLogUniform
seems to be a uniform distribution instead of log-uniform. Although samples from BiSymmetricLogUniform.samples() are distributed as BiSymmetricLogUniform, the NUTS samples are not, see plots below. The key to makeBiSymmetricLogUniform
working for NUTS is to correctly implement thelog_prob
method, andicdf
for nested sampling ofjaxns
.