lichess.org

Stockfish's "opinion" on balancing horde chess

To provide some data for the discussion, in the following I list results of multi-variant Stockfish playing itself at different time controls. The base time ranges from 10 to 180 seconds (with increment scaling in the same way) to see how the score evolves with increasing time control, i.e., increasing playing strength.

I have used the standard horde setup and two possible alternatives with the same number of pawns. For each setup and time control, I give the score in % from white's (i.e., the horde's) perspective. I have already mentioned parts of the results in another thread, but I have added a few new results here.

standard
rnbqkbnr/pppppppp/8/1PP2PP1/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPPPPPPP w kq - 0 1
----------------------------------
10"+0.1"
28.1%

30"+0.3"
23.9%

90"+0.9"
19.2%

advanced
rnbqkbnr/pppppppp/8/1PPPPPP1/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPP2PPP w kq - 0 1
----------------------------------
10"+0.1"
64.1%

30"+0.3"
60.9%

90"+0.9"
64.1%

180"+1.8"
61.3%

centered
rnbqkbnr/pppppppp/8/2PPPP2/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPPPPPPP w kq - 0 1
----------------------------------
10"+0.1"
25.4%

30"+0.3"
20.0%

90"+0.9"
21.8%

Wow that's quite a big score for advanced...Do you think the match today will be very unbalanced? Maybe switch to advanced instead of my idea?

@FischyVishy
The data suggests this, but human games with the standard horde setup already show quite different results compared to Stockfish's results, so I am not sure whether this is a good indicator for results of human games.

Even at high levels the score in human games seems to be much closer to 50% than for Stockfish in self-play. This suggests that in human games white's score usually is higher, so I would guess that white's score with the advanced setup would be >60%. Based on this I personally would vote for the advanced setup, but finally the players and you have to decide.

I have not tested your suggestion with Stockfish yet since the advanced setup already resulted in a white advantage, but if the results of the match today show that this setup is balanced for human players, I will certainly study it with Stockfish.

Thanks for the explanation.

Yeah, let's see what the match today yields.

@FischyVishy
In the meantime I have run tests with the additional e5 and d5 pawns. I expected it to be one-sided, but the results are really extreme:

38 pawns
rnbqkbnr/pppppppp/8/1PPPPPP1/PPPPPPPP/PPPPPPPP/PPPPPPPP/PPPPPPPP w kq - 0 1
----------------------------------
10"+0.1"
96.2%

30"+0.3"
97.5%

Ha that's crazy...Let's see how it fares with humans.

lichess.org/xjRl3hskTgkD @FischyVishy this is a bit like chess+horde white is 1point material up the comp says its .0.7 for black but it must be completly winning for white totally not practical playable for black this is not the same as horde but the comps evals about this things with the pawn is not always right.

@opperwezen Interesting, but this looks like an extreme case. I do remember you playing this stuff with jokido. I wonder if @ubdip would be willing to run some tests on this position.

jep this is a extreme case but pretty sure about it thats its +-

how can i see your stream

@FischyVishy
I think it does not make much sense to test this with standard chess Stockfish, since it causes undefined behavior in Stockfish if such positions (with too many pieces) are put in, which can hurt playing strength and cause crashes. Furthermore, I suppose that standard chess Stockfish is bad at evaluating horde-like positions, so the games might anyway not be on a high level.

Reconnecting