diff options
author | c-blake <c-blake@users.noreply.github.com> | 2018-12-23 07:23:20 -0500 |
---|---|---|
committer | Dominik Picheta <dominikpicheta@googlemail.com> | 2018-12-23 12:23:20 +0000 |
commit | e1d5356ae9fe0e289cc6fe21fec486c54f89400a (patch) | |
tree | 827a126eedd14656cfc2645ada998d4851be9d3d /nimpretty | |
parent | d407af565f164a2182e9d72fa69d1263cd7b3f9f (diff) | |
download | Nim-e1d5356ae9fe0e289cc6fe21fec486c54f89400a.tar.gz |
Add ability to sample elements from openArray according to a weight array (#10072)
* Add the ability to sample elements from an openArray according to a parallel array of weights/unnormalized probabilities (any sort of histogram, basically). Also add a non-thread safe version for convenience. * Address Araq comments on https://github.com/nim-lang/Nim/pull/10072 * import at top of file and space after '#'. * Put in a check for non-zero total weight. * Clarify constraint on `w`. * Rename `rand(openArray[T])` to `sample(openArray[T])` to `sample`, deprecating old name and name new (openArray[T], openArray[U]) variants `sample`. * Rename caller-provided state version of rand(openArray[T]) and also clean up doc comments. * Add test for new non-uniform array sampler. 3 sd bound makes it 99% likely that it will still pass in the future if the random number generator changes. We cannot both have a tight bound to check distribution *and* loose check to ensure resilience to RNG changes. (We cannot *guarantee* resilience, anyway. There's always a small chance any test hits a legitimate random fluctuation.)
Diffstat (limited to 'nimpretty')
0 files changed, 0 insertions, 0 deletions