Free online Chess server. Play Chess now in a clean interface. No registration, no ads, no plugin required. Play Chess with the computer, friends or random opponents.
Sign in
Reconnecting
  1.  Improving Horde
  2. Forum
  3. SF Horde Strength

Alright I played alot with SF today.
To beat it with white wasn't to challenging (unless you don't blunder :P).
It just makes big opening mistakes and always alows black to breakthrough quite easily.
For Example:
lichess.org/kvbeJBdL/black

With white tho it was unbeatable. At least for me and for today...
I played like 35 games with white today and got crushed every time.

@Stubenfisch
Very interesting, thanks. I think it is the first time that we have a decent number of games at a reasonable time control between Stockfish and a top horde player. The results seem to confirm that SF's self-play results (lichess.org/forum/team-improving-horde/stockfishs-opinion-on-balancing-horde-chess ) are a good estimate of white's and black's chances in horde chess. Or maybe it only confirms that SF does not know how to play well as white, neither against itself nor against human players?!

It still is very strong with white. You really need to watch out as it is extremly resourceful and tricky.
SF main weaknesses are in the opening because it allows you to break through quite easily with a knight sac on h4/a4
For Example in this game:
lichess.org/MBXz9cen/black#0

Maybe it simply doesn't see a really good way to respond to this threat as the lines humans invented to prevent Nxa4/xh4 ideas mostly lead in losing positions for white which are quite difficult to handle for human players.
For example:
lichess.org/g8hYt3AH/black#0

@lecw
Two of your suggested changes (regarding connected and pawn-shelter bonuses) performed well and were merged into the master branch:
github.com/ddugovic/Stockfish/commit/983ef6b4bfd224f07ace9af385970e97a55ea5b1
github.com/ddugovic/Stockfish/commit/9420ed7fcce72401ac0c53bca2b878073cd7bd9a

The Elo gain from these two patches for horde chess should in total be around 10 Elo. Thanks for your contributions.

After over 100 4+4 Games today with alot of frustraition and rage I finally managed to beat SF8 with white!

lichess.org/QLYmqDtN/white

I took two takebacks tho :/

Yay, congrats ^^ (Holy sh*t you're 2700 now ? gg)
So does black just has better chances than white from the starting position, given all we currently know ? Or is just SF stronger with black than white ?

Thanks ubdip :) I'll come back to you if I have any more implementable ideas.

I think black just has an advantage from the starting position. And SF is better with black because of this advantage.
@lecw Thank you :) I am overrated tho I just farmed realy hard lately :3

With the white pieces, neither SF itself nor Stubenfisch seem to be able to score close to 50% against SF. Since it is hard to find much better human players to test that further, one idea might be to run SF self-play games with a bias, i.e., the side playing white gets more thinking time in order to simulate a stronger white player.

I think scoreing ~50% is close to impossebile, because blacks advantage just seems to big.
But I like your Idea with the time @ubdip !

I ran some tests where white had 2x or 4x more time (see http://35.161.250.236:6543/tests/variant/horde). It is only 500 games per time control, but the trend is relatively clear. Black's score seems to improve when increasing the time control but keeping the same ratio (e.g., moving from 10s vs. 5s to 20s vs. 10s). Since this can not be explained by a higher draw rate (which is negligible in horde), it indicates that even if white is the stronger player, black seems to have an advantage if playing at a high level.

Even with only 1/4 of the time, which corresponds to a 200-250 Elo handicap, black can score around 50% at sufficiently long time controls whereas black clearly loses at shorter time controls. Testing this at even longer time controls is not really feasible, but I would expect the score to go up further.

This topic has been archived and can no longer be replied to.