Nice work! Thanks! Worth trying!
Each loop iteration in fast_spread() adds the new pairwise penalty (weights) straight onto the running ranks total, without dividing by how many states have been picked so far, unlike the original fast package.
As a result, when picking 3+ states in a round, the accumulated penalty can grow larger than the FAST score itself (which is bounded to [0,1] + alpha*[0,1]), becoming the dominant factor in choosing the remaining states.
Is this the intended behavior/change?
Nice work! Thanks! Worth trying!
Each loop iteration in fast_spread() adds the new pairwise penalty (weights) straight onto the running ranks total, without dividing by how many states have been picked so far, unlike the original fast package.
As a result, when picking 3+ states in a round, the accumulated penalty can grow larger than the FAST score itself (which is bounded to [0,1] + alpha*[0,1]), becoming the dominant factor in choosing the remaining states.
Is this the intended behavior/change?