search

bmizerany / perks

Effective Computation of Things
BSD 2-Clause "Simplified" License
186 stars 53 forks source link

Fix error calculation for tests. #11

Closed beorn7 closed 9 years ago

beorn7 commented 10 years ago

Change sample generation to rand.NormFloat64 to expose previous problem.

(The current way of error calculation doesn't really work when sample values are close to zero. In my understanding, ε is meant to be the error of the quantile rank, not of the resulting value.)

beorn7 commented 10 years ago

@bmizerany Didn't mention you on this one... :)

beorn7 commented 9 years ago

I'll close this. Found a few other issues. If you are interested in my (still quite messy) fixes, check out my fork at https://github.com/beorn7/perks (in particular https://github.com/beorn7/perks/commit/4a53c8d88a15e9591efbc4204d2c0db64a1f2db1 ).