Closed bmckown closed 9 years ago
you won the jackpot @bmckown
When choosing the bounds I ran the tests 1000 times. And was good (running few times). So we expect the failure probability to be < 0.001 (+/- some bias). Last results of 32 when current bound is 20 indicates that the distribution of results of the algorithm may have a high variance.
@pankajgupta, @bmckown we can: 1) increse the bounds (even further?, would that help?) 2) disable the test 3) implement sth like: when fails try again only once Any thoughts?
Ha. Whats funny is this is actually the second time I've seen the tests fail in the same spot--once it failed locally and now in the travis-ci build.
Is part of this test to actually test the random graph generation? Could we test the triangle counting algo on a static graph of our choosing?
No. It should test randomized triangle counting algorithm.
It needs a big graph to have good precision.
Maybe we should store sam random graph(s) to a file and use this one. Now we have two randomizations.
Agree with you guys that we need to make this robust. @szymonm Is it possible to do that by choosing seed values of all random variables in the triangle counting spec and running it on a pre-stored graph
I am also getting this test to fail unpredictably (e.g., in the travis ci build of https://github.com/twitter/cassovary/pull/152) . @szymonm -- would be great to make this more predictable sooner.
Please check this out:- TriangleCount failed. https://travis-ci.org/twitter/cassovary/jobs/51226753
That is right. #161 is a fix
The randomly generated graph does not consistently provide a triangle count of 0 +/- 20.
https://travis-ci.org/twitter/cassovary/jobs/48922273#L1352