Virtual Loss Apply

Problem

Apply a virtual loss to a node so parallel MCTS workers spread out.

Return [[w - virtual_loss for w in wins], [v + 1 for v in visits]], discouraging other workers from selecting the same path.

Hint

The virtual visit count also lowers the node's average value.

case 1

in: [[3,1],[2,4],1]

expected: [[2,0],[3,5]]

case 2

in: [[0],[0],0.5]

expected: [[-0.5],[1]]

case 3

in: [[],[],2]

expected: [[],[]]

case 4

in: [[2,-1],[5,2],0.25]

expected: [[1.75,-1.25],[6,3]]

Starter codePython

def virtual_loss_apply(wins, visits, virtual_loss):
    # Your code here
    pass

4 test cases · run your code against them in the editor