Closed dominusmi closed 4 years ago
The equation is right as it is. If the other walkers' virtual reward were +30% better than yours, you would clone with probability 0.3, but if it were more than double of yours, then it is +100% better and then you will just do clone, without flipping any coin. It is the same case where his reward is lower and you get a negative prob. of cloning, just meaning you won't clone.
After all, you get a "clone prob." P and a random number R in [0, 1] and the code just clone if (R<P), no problem if P is whatever real number you get, it always makes sense, but some people feel uncomfortable with this. Just clip probabilities into [0, 1] to recover peace of mind!
Alright thank you for the answer :)
Hello!
I believe there might be an error in the cloning probability, both in the paper and the code.
I think the third equation should be
, since we know ![](https://latex.codecogs.com/gif.latex?VR_k>VR_i)
With your current probability, it will very likely be greater than 1, for instance if
will give probability = 1.
On the other hand, using the
as denominator, we get p=0.5.
Was this an error, or was it an intentional choice to bias the probability upwards, and if so why?