-
Notifications
You must be signed in to change notification settings - Fork 224
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
Noncentral F distribution CDF edge cases #1198
Comments
Reproduced, first one is an easy fix. Not so sure about the second yet... |
Update: working on this now, there are some easy to fix numerical-stability issues which sort of fix both, but ideally we need some asymptotic expansions for the incomplete beta with large parameters: these exist, and I have them somewhat working, but they are deeply numerically unstable, especially near the saddle point. Hopefully more soon... |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
SciPy is thanks to Boost finally able to remove parts of the ancient cdflib Fortran library. Recently, the noncentral F distribution CDF was fully replaced by the boost implementation in scipy/scipy#21505, resulting in a far more accurate results for a wide range of inputs. Still, edge cases were uncovered which we would like to share with you. They occur for very large and very tiny values for the degrees of freedom. Note that for the second case even Mathematica is not able to compute a result, so maybe we do not need to thrive for a feasible value but raise an error or return nan. @mdhaber found a limiting formula that might be helpful: scipy/scipy#21505 (comment)
The text was updated successfully, but these errors were encountered: