Monte Carlo Data

[AdminX]

Monte Carlo Data

[d]8/5k1p/5q2/8/1R6/6P1/5P2/6K1 b - - 0 0



Should not the Draw rate be higher here in the above position?

[Rubinus]

Probe random moves - 19% for white
After Rf4 and King moves (or Rxh4) remis ever ...
Monte Carlo is only toy.

Monte Carlo, R3, depth 9, start position

Code: Select all

1.e3  20 65.0%
1.c4  20 62.5%
1.h3  20 60.0%
1.d4  20 55.0%
1.Nc3 20 55.0%
1.Nf3 20 52.5%
1.g3  20 45.0%
1.a3  20 45.0%
1.d3  20 37.5%

Whitout my like 1.e4! New Rybka have some aversion

[AdminX]

Yes, funny how if I make white to move there is no problem.
[d]8/5k1p/5q2/8/1R6/6P1/5P2/6K1 w - - 0 0

[Dr.Wael Deeb]

I don't like this feature and most probably won't use it....

[Ovyron]

It was useful to reveal that white has no winning chances, at least.

[shiv]

In this case, the fortress plan is hard for the computer to work out if it does not have access to tablebases.

[AdminX]

In this case, the fortress plan is hard for the computer to work out if it does not have access to tablebases.

It did have acess to Tablebases.

[shiv]

Sorry for the late reply. 7 man tablebases are needed for the right solution in this case and I doubt you had 7 man tablebases installed

If the Monte Carlo score is wrong even with 7 man tablebases, then we have a major bug.

[chrisw]

Monte Carlo Data

[d]8/5k1p/5q2/8/1R6/6P1/5P2/6K1 b - - 0 0



Should not the Draw rate be higher here in the above position?

This interests me.

I always thought Monte Carlo technique wouldn't apply in chess (more so for learning from a no-knowledge base) because so many games would end up shuffle shuffle aimlessly and be drawn, but take way too many moves to detect. ie too many games wouldn't terminate.

Monte Carlo is a way to learn for Backgammon, very effective, it produces a GM strength program after 500,000 games or so via neural net, I did that for backgammon some years ago, but ran into the never ending games problem when similar idea tried for chess. (Backgammon essentially is a one-way street, eventually all pieces exit the board and game is over by definition).

Do you notice this game incomplete problem for the Rybka Monte Carlo-ing? Is it mentioned in the readme, or?

[Anil]

Chess also has some exit conditions apart from Mate and Stalemate:
1. Draw due to 3-move repetition
2. Draw due to 50 move rule

I guess Rybka Monte Carlo analysis would be using these exit clauses to end the games quicker.

But, there may be some games that may go very long, in case all options/moves are tried.