I noticed that the Neural Net Test caused the build to fail. There is nothing wrong in the code, but unfortunately the algorithm is non-deterministic so there is always the chance it will fail, if we are very unlucky.
What should we do about these kinds of tests? I would like to be testing them in case we break something in the implementation, but there are times where this backfires and unluckily the build fails. Another solution I can think of is increasing the threshold for error, but this may also backfire (we make a change that makes the implementation worse, but we don't notice it because it doesn't surpass the threshold).
I think the best way to go is increase the threshold by a little (say, 0.33), at least for now.
I noticed that the Neural Net Test caused the build to fail. There is nothing wrong in the code, but unfortunately the algorithm is non-deterministic so there is always the chance it will fail, if we are very unlucky.
What should we do about these kinds of tests? I would like to be testing them in case we break something in the implementation, but there are times where this backfires and unluckily the build fails. Another solution I can think of is increasing the threshold for error, but this may also backfire (we make a change that makes the implementation worse, but we don't notice it because it doesn't surpass the threshold).
I think the best way to go is increase the threshold by a little (say, 0.33), at least for now.