08-28-2022, 09:02 AM
In the case of Alpha-Beta, the visualisations don't mean much at all. I did at one point try to make it visualise something meaningful (like, trying to estimate how much "search effort" the algorithm is putting into a move based on the search depth or something like that), but I've found this difficult for Alpha-Beta due to it's relatively non-smooth / discrete / jittery behaviour during the search, things like pruning, and the fact that for certain situations it only obtains upper or lower bounds on values, instead of actual value estimates.
In comparison, MCTS-based algorithms are much "smoother" and have two very nice and meaningful quantities that can be really easily and smoothly visualised through size and colour: visit counts and value estimates.
In comparison, MCTS-based algorithms are much "smoother" and have two very nice and meaningful quantities that can be really easily and smoothly visualised through size and colour: visit counts and value estimates.