Open
Description
Description
For arithmetic stability in CDF calculations, we should have an approximate Phi
and its complement on the log scale.
Note
- Ideally this will be done with custom derivatives
- Ideally this will be properly vectorized
If not, there's a log_inv_logit
that could be called.
A naming convention for complements needs to be established. logc_Phi_approx
seems a bit too cryptic despite our having lcdf
and lccdf
functions.
Current Math Version
v2.18.0