Hi Daniel, I don't really see why you declare this failure as "serious". From the source (and the bug name), it basically means that theano took a bit longer to calculate than scipy for the same calculation. This may depend on the scipy version and optimization, and also on the timing (load) during the test -- it is calendar time what is taken here.
IMO, the failure is severity "normal", maybe you could consider resetting the severity to this value? Best regards Ole
