-
Notifications
You must be signed in to change notification settings - Fork 33
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fixes to correlation bijectors #301
Comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The current bijectors for correlation matrices and their Cholesky factors when used in Turing are still quite brittle. A model just drawing from a uniform
LKJCholesky
regularly errors:As the error message shows, this is coming from hitting the method
Bijectors.jl/src/interface.jl
Lines 213 to 216 in 2402be2
Bijectors.jl/src/bijectors/corr.jl
Line 319 in 2402be2
At the same time, I would like to implement all
with_logabsdet_jacobian
methods for these bijectors, as the logdetjacs are trivially computed as part of the transform.Likewise, there are some naming issues here. A number of the internal functions have
lkj
in the name. LKJ is one distribution of correlation matrices, but it's not the only one. This should be changed tocorr
orcorrchol
. Similarly,VecCholeskyBijector
is incorrectly named. This bijector does not map from any Cholesky factor to a vector. Rather, it maps only from Cholesky factors of correlation matrices, so it should be renamed toVecCorrCholeskyBijector
.The text was updated successfully, but these errors were encountered: