cdani wrote:Another idea is to play human + komodo without a knight vs human. The human point of view can be interesting to find complications than he will know that can't be well handled by the sole human.
Yes, I talked with Victor about this idea just after the last game.
Sometimes, I was looking at Komodo changing its mind over the depths and thinking "this move seems bad, exchanging pieces with no good reason or too passive or too closed position...". But I can't see everything Komodo sees. May be I would have done worse than Komodo alone. Victor seems to me in great shape last Sunday !
lkaufman wrote:If Victor also got the white pieces that might be about right; with the Black pieces and f7 missing it would be too difficult for him I'm sure, as the missing f7 pawn forces Black to make serious opening concessions. Or with Victor playing Black, maybe knight for the "c" pawn which is a much smaller problem for Black than the missing "f" pawn. Or perhaps alternating colors with knight for "c" pawn. Comments, anyone?
It can be estimated based on engine self-play at time handicap adjusted to material handicap. I should have done this earlier, but it seems that knight is a bit too much for a 2200 player, assuming Komodo is 3200 FIDE. Previously I concluded that 2 pawns (c2 and f2) handicap is worth 600 ELO points at slow TC (400 at ultra-fast), today I saw that knight handicap is 800 ELO points at ultra-fast (8 draws in 400 games), meaning 1200 ELO points at slow TC. That would mean a performance of Komodo of 2000 FIDE without a knight, assuming the regular opening Komodo is 3200.
Is this knight versus pawn at draw odds too? Seems too hard for a 2200 human in this case.
Ferdy wrote:Why Komodo will not take the pawn in a4? Seems to me there is something wrong with contempt.
R7/5p2/5k2/6b1/p1KP4/7P/1r6/8 w - - 0 56
I don't think it has to do with contempt. Much more likely is that it has to do with EGTB. Maybe we shouldn't even use them in these handicap matches, or at least not until the tablebase position is on the board. If we're winning we will probably win anyway against a much weaker opponent without playing perfectly, while if we are losing we should not worry about playing theoretically best moves. I'll have to check into this.
Yes, I think EGTB is currently not very well suited to play against a piece odd. Komodo should try to swindle more or simply disable EGTB when more than 6 (or 7) pieces are on the board. In round 2, the same thing happened : 68.f4??, probably try to avoid to enter into an EGTB lost position.
Jesse Gersenson wrote:What was contempt set at this week and last week?
I noticed that Komodo played passively against me (giving rook odds) if contempt was set too high.
Yes, for this reason Contempt was reduced from 150 last weekend to 100 this weekend. I think one reason Komodo did worse this time and seemed more passive is simply that I used the best opening moves the first weekend, and had to resort to three inferior opening moves this time to avoid repeat openings. I suppose if the only goal was to win I would have repeated 1d4 and perhaps 1Nf3 (which are likely to transpose), but getting variety for the spectators and avoiding possible preparation were also considerations. I'm not yet sure how we'll deal with this issue in the next (two pawn handicap) match.
Hi Larry,
This looks to be a serious flaw in the system you have adopted. You can't afford to jeopardize the final result for the sake of "variety" and avoiding preparation - the whole thing seems to be at cross-purposes and, I'm afraid, it detracts from the spectator's experience and diminishes the whole experiment you are running.
Whatever system you adopt in the future, it will be much better if the human does not know what openings he will have to face, thus being forced to play on his own - and the same will go for the engine, presumably.
On the plus side, I do like the fact that Komodo now has a superior contempt function, something end users will probably look forward to.
Jesse Gersenson wrote:What was contempt set at this week and last week?
I noticed that Komodo played passively against me (giving rook odds) if contempt was set too high.
Yes, for this reason Contempt was reduced from 150 last weekend to 100 this weekend. I think one reason Komodo did worse this time and seemed more passive is simply that I used the best opening moves the first weekend, and had to resort to three inferior opening moves this time to avoid repeat openings. I suppose if the only goal was to win I would have repeated 1d4 and perhaps 1Nf3 (which are likely to transpose), but getting variety for the spectators and avoiding possible preparation were also considerations. I'm not yet sure how we'll deal with this issue in the next (two pawn handicap) match.
Hi Larry,
This looks to be a serious flaw in the system you have adopted. You can't afford to jeopardize the final result for the sake of "variety" and avoiding preparation - the whole thing seems to be at cross-purposes and, I'm afraid, it detracts from the spectator's experience and diminishes the whole experiment you are running.
Whatever system you adopt in the future, it will be much better if the human does not know what openings he will have to face, thus being forced to play on his own - and the same will go for the engine, presumably.
On the plus side, I do like the fact that Komodo now has a superior contempt function, something end users will probably look forward to.
Regards,
CL
The problem is that if we always play the "best" move the human will learn from each game and improve his play. But you are right, in this match playing inferior first moves was too costly. I guess I'll have to make small handicap books that provide variety for the first three moves or so wihout giving up too many centiply in expected score.
