Post by alandau on Jun 9, 2014 23:29:56 GMT -8
I noticed in a recent match of Three-Player Connect Four that my player was basically making all its decisions based on what would best feed the player after me. Its expected score was around 49 for its favored move and significantly lower for its other options. (This, despite that the player before me was Random and had played into a poor position to mount any kind of threat.)
I suspect that this is because 3pConnectFour gives scores of 0 to each player in the event of a draw. This isn't usual practice; in this case, it makes any random found path to (partial) victory look much more appealing than the random space of depth charges.
I'd make the change to give 50 to each player in a draw, but I'm concerned that this would lead to lots of stalemates, like what happened to Qyshinsu after the same change was made there. Another possibility is giving the "assisting" role a reward that's still better than a draw, say 100/50/0 vs. 33/33/33, or 100/68/0 vs. 56/56/56 if you want to keep the sum fixed. This might maintain the incentive to push the game in the next player's favor, while doing less to discourage players from acting in their own interests.
Any thoughts?
I suspect that this is because 3pConnectFour gives scores of 0 to each player in the event of a draw. This isn't usual practice; in this case, it makes any random found path to (partial) victory look much more appealing than the random space of depth charges.
I'd make the change to give 50 to each player in a draw, but I'm concerned that this would lead to lots of stalemates, like what happened to Qyshinsu after the same change was made there. Another possibility is giving the "assisting" role a reward that's still better than a draw, say 100/50/0 vs. 33/33/33, or 100/68/0 vs. 56/56/56 if you want to keep the sum fixed. This might maintain the incentive to push the game in the next player's favor, while doing less to discourage players from acting in their own interests.
Any thoughts?
