hnb22

Hi, it is my code. If your referring to borjha19, thats an old Github account of mine that I don't use and forgot about. Not sure why its showing up...

Those tests and benchmarks look good. The patch works well, thank you. Why is 15 the conditional check for cancellation by the way? Also, do you want me to apply...

Here's some benchmarks for new code, higher precision refers to AGM method: | Input | Precision | Production Mean | PR Mean | Speedup/Slowdown | |----------------|-----------|-----------------|----------|-------------------------| | log(1.001) | 53...

Here's some benchmarks with new higher 10000 check ```python import pyperf from mpmath import log, mpf, workprec runner = pyperf.Runner() x = 1.00001 # cancellation = 17 precision_levels = [...

Yeah, it's slower... Would you prefer the PR stays open or can I close it? Don't know how to proceed. Sorry, thanks for your time on this.

Yeah, it's slower... Would you prefer the PR stays open or can I close it? Don't know how to proceed. Sorry, thanks for your time on this.

I'll continue to work on it. Give me a couple days, I'll provide an update. > Think a bit about the asymptotics here: with gmpy, AGM is O~(prec) while Taylor...

## Code ```python import pyperf from mpmath import log, mpf, workprec runner = pyperf.Runner() for x in [1.001, 0.999, 1.0000000000001, 0.999999999999]: for prec in [2000, 500, 750]: with workprec(prec): m...

I'm not sure I'm convinced log2(wp) is precise enough, but offers a heuristic to ensure taylor is optimal. Because line 696 increments wp by cancellation, cancellation will never be greater...

Hi, I updated PR title and opening comment/description.