search

LambdaConglomerate / x9115lam

2 stars 0 forks source link

cdom or Golinski Bug #35

Closed aisobran closed 8 years ago

aisobran commented 8 years ago

I get this with Golinski:

losses = [math.exp((a - b)/n) for (a,b) in zip(c1, c2)]
OverflowError: math range error
aisobran commented 8 years ago

Get the same error with Water

aisobran commented 8 years ago

Also with DTLZ3. This is either bug with cdom or with these three models.

meneal commented 8 years ago

In grapher why is line 24: if len(s.listOfVectors[1][1]) <= 2: and not if len(s.listOfVectors[1][0]) <= 2: ?

I'm looking at that DTLZ bug, but I'm getting list index out of range errors in grapher before I hit the overflow.

aisobran commented 8 years ago

You're right if there is only one value in the frontier it should be zero.

I'll push a fix with the added true pf support.

meneal commented 8 years ago

I'd say right off the bat that it looks like we're stuck with normalizing for at least these models. I really tried to get away from that mostly because I was having so much trouble getting cdom to work correctly.

The only other possibility would be to drop these three models. But I don't think that's a great option. Once we have normalization rigged back in, I'll figure out how to either use normalization for just these three models or for all of the models.