This is a tentative fix to the BIC criterion for ellipsoid splitting introduced in #286
There a factor of two was skipped in the calculation making the splitting of ellipsoids harder.
Apparently that leads to significant slowness observed in #440
I'm a bit torn by this issue and whether it's right change. One wouldn't want the results to be that sensitive to that factor of two here and I'm concerned if this will break some other cases...
This is a tentative fix to the BIC criterion for ellipsoid splitting introduced in #286 There a factor of two was skipped in the calculation making the splitting of ellipsoids harder. Apparently that leads to significant slowness observed in #440
I'm a bit torn by this issue and whether it's right change. One wouldn't want the results to be that sensitive to that factor of two here and I'm concerned if this will break some other cases...
One thing that this change trips is the test_ellipsoids test case https://github.com/joshspeagle/dynesty/blob/d9164354421f4a010ee2bc232cb4eda2f4f90a88/tests/test_ellipsoid.py#L265 where I check if the number of clusters in 6 x 6 x 6 x 6 grid in 4d is within 10% of 6^4. I don't know if it's a serious concern or not...
Thoughts ?