Playing a (non)perfect opponent Einstein forum
4 replies. Last post: 2006-11-14Reply to this topic Return to forum
hanfried_c at 2006-11-09
A long time ago (in march 06) Theo van der Storm wrote in an other topic:
“You … have the option to … adapt your strategy to the playing style of the opponent. I don't do that. I pretend to play an all-knowing opponent. And you?
To stir up a discussion I would say:
If you are good enough you don't waist energy to study your opponents.”
After a few dozen games as the operator of the computer program hanfried_c against (non)perfect opponents I can't agree with Theo.
Hanfried does not say “this is the best move” but he gives estimated winning probabilties for all legal moves.
My general procedure is to feed hanfried with the current position, to compare the chances, and to execute the “best” move. But sometimes there is a tie (hanfried gives only two digits, for instance 68%). In many cases it is clear that both moves are of absolutly equal quality. So it does not matter, which move I execute.
But in some cases there are absolutely different moves of (nearly) same quality. I started to analyze such moves and found out that moves of equal qualty against a perfect player can perform quite different against a (non)perfect player. I want to illustrate this with an extreme example.
Assume that you are a perfect player of the well balanced 2-person-game A, so you will win with a probability of 50% against an other perfect player. Now you have the choice, to play a game A against a perfect player or to flip a coin. In both cases you will win with prob. 50%.
But playing against a non-perfect player things are different. The chance to win the coin-flip is still 50%. But the chance to win a game A against this non-perfect player is higher than 50%. So it is better to play game A than to flip a coin.
Of course hanfried isn't a perfect player but in the argumentation you can replace the two perfect players by two players of equal strength and the non-perfect player by a weaker player.
Where is the connection to EinStein?
In my opinion the following strategy is promising.
If you have two (or more) moves of equal quality and you think you are stronger than your opponent than try to slow down the game. Give your opponent the chance to make mistakes (or at least to make more and greater mistakes than you). Avoid subgames where your opponent has an obvious strategy.
On the opposite if you are feeling weaker than your opponent, try to force the game. Try to come to a quick end. Don't give your opponent the chance to “outact” yourself. Avoid subgames you poorly understand.
Theo van der Storm at 2006-11-10
You took some time to write your well formulated reaction.
Maybe you can't agree with me, but I can agree with you fully :-)
We were just setting different goals.
My goal was to create the strongest EWN player.
Your goal is get the best score in a mixed competition.
It may come as a big surprise to you, that I would even go a step further.
Nowadays the differences in the top EWN programs are probably so small,
that the only way to achieve your goal is to apply opponent modeling.
In other words: Nobody can afford to be lazy and this is becoming a different ball game all together.
12259 x “This is becoming a different ball game”.
Sorry, I meant: “altogether”.
hanfried_c at 2006-11-14
At first: Good to know, that you still read this forum. I hope you will also play a bit in the future (esp. Monthly Cups and Championships).
I needed some time to read the old forum entries. ;-)
The problem is the definition of “strongest EWN player”.
In an infinitely long match the “strongest EWN player” can rely on his/her/its own strength. But we have only finite matches.
My suggested strategy doesn't work well against computer programs. Their strengths and weaknesses are too similar to the strengths and weaknesses of hanfried.
Perhaps it is possible to outsmart fraggle(*). But if Ingo S. is clever enough he switches out his “perfect play”-bonus if he is playing against computers.
(*) I hope to remember right that fraggle was the program witch gave a bonus for positions whose winning-pobabilty he is knowing correctly.
Falsifian at 2006-11-14
Speaking of modeling your opponent, I just made a post about Roshambo (Rock-Paper-Scissors) in the main forum. In roshambo opponent modelling is the _only_ tactic.