If prWeakTrueMiss is too small (right now it is 0.85), we are introducing -Inf
in lines 335/336/338
int nQuHuPt = nodesQuPtBM.andNot(nodesHtBM).andNotCardinality(nodesHfBM);
double cprQuHuPt = Math.pow(prWeakTrueMiss, nQuHuPt);
In my code, nQuHuPt in use with HPO becomes larger than 10,000. Then cprQuHuPt = Math.pow(0.85, 10000) is 0 and in line 338 we compute Math.log(cprQuHuPt).
Questions:
Do you get similar values for nQuHuPt? (I have re-implemented your code, so not sure the problem is on my side)
If yes, any thoughts on this? Increase prWeakTrueMiss? But then we also have to significantly decrease prWeakFalsePositive, because in line 329 we calculate Math.pow(1-(prWeakFalsePositive + prWeakTrueMiss), nQfHuPt).
If
prWeakTrueMiss
is too small (right now it is 0.85), we are introducing-Inf
in lines 335/336/338In my code,
nQuHuPt
in use with HPO becomes larger than 10,000. ThencprQuHuPt = Math.pow(0.85, 10000)
is 0 and in line 338 we computeMath.log(cprQuHuPt)
.Questions:
Do you get similar values for
nQuHuPt
? (I have re-implemented your code, so not sure the problem is on my side)If yes, any thoughts on this? Increase
prWeakTrueMiss
? But then we also have to significantly decreaseprWeakFalsePositive
, because in line 329 we calculateMath.pow(1-(prWeakFalsePositive + prWeakTrueMiss), nQfHuPt)
.Any help greatly appreciated.