simplify mcse_elpd using log-normal approximation #246
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Years ago I made mcse_elpd using a simple quadrature, but now switched to use a more efficient log-normal approximation. Given the mean and variance in linear scale, and knowing that the logarithm is closer to normal, we approximate the uncertainty in linear scale with log-normal and compute the corresponding normal distribution parameters in log scale using the equations shown, e.g., at https://en.wikipedia.org/wiki/Log-normal_distribution#Arithmetic_moments
The change makes relative change of 0.01%-0.1% compared to stored reference values and thus some reference rds files were updated.