This PR adds entropy implementations, using scipy.stats distributions as the reference implementation for tests where possible. There is some overlap with #1706.
Currently, only one entropy implementation is available (for the LowRankMultivariateNormal). It might be useful to switch entropy to use lazy_property as for mean and variance. That would make the interface consistent (entropy is a property of the distribution like mean or variance), and it would allow caching if the evaluation is more involved. What do you think, @fehiepsi?
This PR adds entropy implementations, using
scipy.stats
distributions as the reference implementation for tests where possible. There is some overlap with #1706.Currently, only one
entropy
implementation is available (for theLowRankMultivariateNormal
). It might be useful to switchentropy
to uselazy_property
as formean
andvariance
. That would make the interface consistent (entropy is a property of the distribution like mean or variance), and it would allow caching if the evaluation is more involved. What do you think, @fehiepsi?