Closed cmgoold closed 1 year ago
std_normal_lcdf(0)
is constant, so it doesn't affect the importance ratios nor the sensitivity measures. If you like to include that in your code, it doesn't harm anything.
Thanks @avehtari. Testing a few examples indicated that was the case but wanted to make sure it was in general.
Do we care about the exact parameter bounds/additive constants when computing the log joint priors for powerscaling? In the examples, this doesn't seem to be the case, as the computation of the log priors don't account for the parameter bounds. For instance, for a scale parameter, the examples state something to the effect of:
but this really should be:
if we care about the precise likelihood contribution.
@n-kall @avehtari I couldn't find this in the original paper, but would be interested in your thoughts. Thanks!