Closed mdhaber closed 1 year ago
I did some benchmarking as in https://github.com/scipy/scipy/issues/7242#issuecomment-1365495412. To invert argus.cdf
, _chandrupatla
takes about twice as long as brent
for scalars (~3ms VS ~1.5 ms), but consequently, it starts to become faster for arrays with more than one element. For large arrays, it can be over 350x faster.
For comparison, newton
is about as fast as brent
for scalars, and twice as fast as _chandrupatla
for large arrays, but it's unreliable. For the problem above and with a guess of 0.5
, it fails to converge in about 5% of cases.
Work on scipy/scipy#17719. Also completes functionality, documentation, and tests.