Closed FRosner closed 10 years ago
Failure still happening. There seems to be a different problem.
The problem is that 0.00000004 as random value gets treated as 0 and thus generating a discrete feature value 0 with cumulative probabilities (0,0,1), because 0 = 0.
Integration test failed.
If you sample from a distribution with probabilities (0,0,1) e.g., sometimes 0.00000005 gets treated as 0 and the first value gets sampled. This can be avoided by comparing with maximum precision.