I like the way you're thinking, but it's important that the tests be deterministic and repeatable. Otherwise, you're left judging what is "out of regular." Make sure whatever Poisson generator you use can put out repeatable values.
However, we've closed V3.7 to new features. We're trying to digest the ones we have! Work off of a branch of development. Call it 'spike'. Keep it reasonably up to date. Then we can easily pull in your changes once we get this thing out. -tk On Sat, Jan 21, 2017 at 3:46 AM, Darryn Capes-Davis < [email protected]> wrote: > Hi All, > > I have taken a tangent today and have created the following soon to be > submitted as a PR. > > 1. Simulator driver spike generation. > 2. StdQC Spike detection code. > > The simulator driver spike generation code uses random timings of a > Poisson distribution. See http://bit.ly/1wNxBfg for my inspiration. By > default I have coded an average of 4 spikes across the simulated > observation period. > > I have been able to use the test_sim unit test, include StdCalibrate and > StdQC, change the almost equal Unit Test check to 1 decimal place and it > has worked most times. However this does not feel good enough for me. > > So I ask you wise folk, any ideas on a good unit test in this case? I am > thinking to run a simulator with spikes and MinMax and Spike QC set, and > then check that the Archive records do not contain anything out of > regular simulated bounds and the daily summaries do not contain anything > out of regular simulated bounds. Anything else needed? > > Cheers > > Darryn > >
