Job Failed #604163
Job #604163 failed for a48cc01b:
In this commit I had changed the default network for training from MLP to SPIDNA1 and it broke this test (apparently only when run with CUDA). The purpose of this test is to ensure that the same training, when run twice in a row with the same random seed, returns the same result. It worked consistently with MLP but not with SPIDNA1, so there is apparently some random element, at least when run with CUDA, that is not consistent (even though the results are still almost the same within a few decimal places.
For the purposes of fixing the test temporarily I will change it back to MLP but this is still a bug that should be investigated.