What do you think the value of the Queen should be ?

[lkaufman]

[quote="JJJ"]Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?[/quote

I think higher elo is not enough; we would have to come up with some way to make Komodo set more difficult problems for the human, some much better version of Contempt.

[lkaufman]

Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?

I decided to change it a little this is the new position with white to play, what is your prediction this time without looking at any engines? ==> [d]1n1qkbnr/ppppp1pp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQk - 0 1 [d]

This is worse by the bishop pair for Black compared to the previous diagram, so roughly half a pawn worse. Since Black was roughly a pawn better before, he should still be better by perhaps half as much, maybe no longer enough to be theoretically winning.

[carldaman]

Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?[/quote

I think higher elo is not enough; we would have to come up with some way to make Komodo set more difficult problems for the human, some much better version of Contempt.

Hi Larry,

Is a swindle mode a realistic possibility, to enhance the effectiveness of contempt?

I have noticed that tweaking some of the Komodo parameters, including contempt, in a certain way helps Komodo to effectively pose tremendous problems for its opponents, whether engine or human, albeit at the cost of riskier play.

CL

[MikeGL]

Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?[/quote

I think higher elo is not enough; we would have to come up with some way to make Komodo set more difficult problems for the human, some much better version of Contempt.

Hi Larry,

Is a swindle mode a realistic possibility, to enhance the effectiveness of contempt?

I have noticed that tweaking some of the Komodo parameters, including contempt, in a certain way helps Komodo to effectively pose tremendous problems for its opponents, whether engine or human, albeit at the cost of riskier play.

CL

Good point. But also, I guess reducing the time control into blitz or bullet
(lightning) would sway the balance into engines favour. Say a 02:00
minute game, and engines will surely demolish any top 10 GMs with pawn
odds.

Assuming of course that the engine is running on a decent modern hardware.

[lkaufman]

Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?[/quote

I think higher elo is not enough; we would have to come up with some way to make Komodo set more difficult problems for the human, some much better version of Contempt.

Hi Larry,

Is a swindle mode a realistic possibility, to enhance the effectiveness of contempt?

I have noticed that tweaking some of the Komodo parameters, including contempt, in a certain way helps Komodo to effectively pose tremendous problems for its opponents, whether engine or human, albeit at the cost of riskier play.

CL

Good point. But also, I guess reducing the time control into blitz or bullet
(lightning) would sway the balance into engines favour. Say a 02:00
minute game, and engines will surely demolish any top 10 GMs with pawn
odds.

Assuming of course that the engine is running on a decent modern hardware.

No doubt about that. We already demonstrated in two fairly long blitz matches (3' + 2" inc.) that Komodo (on my 24 core) is an even match for two pretty good GMs (Lenderman and Mikhalevski) giving them knight odds. A top ten player should be able to beat Komodo even at blitz with knight odds, but at f7 odds he would have no chance. It is an iron-clad rule that no matter whether the players are computers or humans, the stronger player can give larger handicaps at shorter time limits. We still haven't proven that we can successfully give f7 odds to 2600 level GMs with 90' + 30", the minimum for FIDE rated standard events for high rated players. Perhaps we will soon.

[lkaufman]

Would you try again with knight odds if Komodo was, let's say, something around 3500 FIDE elo ?[/quote

I think higher elo is not enough; we would have to come up with some way to make Komodo set more difficult problems for the human, some much better version of Contempt.

Hi Larry,

Is a swindle mode a realistic possibility, to enhance the effectiveness of contempt?

I have noticed that tweaking some of the Komodo parameters, including contempt, in a certain way helps Komodo to effectively pose tremendous problems for its opponents, whether engine or human, albeit at the cost of riskier play.

CL

There are many potential ways to make Komodo score better against weaker opponents. I expect we will try some before long. Right now we're focused on improving Komodo for the TCEC final, if we qualify.

[yurikvelo]

[pgn]
[Event "Computer chess game"]
[Date "2016.07.27"]
[White "Komodo-10"]
[Black "Komodo-10"]

[SetUp "1"]
[FEN "2bqkbnr/ppp1p1pp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQk - 0 1"]


1. Nf3 Nf6 2. e3 Bf5 3. d3 g6 4. Nc3 *
[/pgn]

D=38, 39 710 MN, Komodo-10
+0,14 4. ... Bg7 5.Be2 0-0 6.h3 c5 7.0-0 a6 8.Bd2 Bc8

[Lyudmil Tsvetkov]

White simply wins.

What is your basis for this claim? I couldn't even tell which side is better, because rooks don't have full value in the opening, and because the king weakness due to f7 missing cannot be exploited without a queen. I used the Monte Carlo feature of Fritz 15 to play it out at ten ply (about IM level I think) and it was dead even after about 140 games. Latest SF gives it about +.4, Komodo 10.1 about -.1, so based on the MC playout Komodo was much closer to the truth.

You tested at very short TC.
I suppose even lower allocation than 1'.
At such TC, actually at any TC lower than 5' a range of eval factors, among those queen value, bishop pair, mobility, but also even psqt and king attacks, tend to overperform to the detriment of more positional eval terms.

I observe that quite clearly, as I am playing a lot of games against top engines with different TC.

[d]2bqkbnr/ppp1p1pp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQk - 0 1

So when Komodo shows 10cps white advantage above (at least my Komodo 8, thanks for the free download), this does not necessarily mean position favours white by only 10 cps. It might favour white by 10cps at 1' per game, but at long TC all the games will be won by white, because white has indeed tremendous advantage above.

My assessment is white is leading by some 110cps, so 100cps higher than what Komodo thinks.
- queen vs N+B gives 150cps queen edge materially
- queen + N = queen vs R+minor in terms of piece complementarity, so those cancel out
- missing d7 pawn=130cps, missing f7 pawn=120cps, so 250cps white edge on this
- central d isolated black pawn gives some 40cps penalty
- compact white pawn structure adds another 20cps for white
- 50cps black edge for quicker development, as is rightly noted, queen develops easier than rooks

So, overall, without the rigth to move, I assess white advantage at some 110cps above.
More than sufficient for a win, as theoretically any mg position with a score over 40cps should be won for the stronger side.

Whe I try playing the position with Komodo 8 with longer TC, white score indeed invariably only rises, with almost no exceptions.

Second position that George posted, with missing only f7 pawn, and black to play, should be a draw, in spite of huge Komodo 8 +110cps black edge score, and indeed LTC game of SF with 1 hour per side only demonstrated that.

In terms of eval for the second position, you do not have isolated d pawn,d pawn not missing, so you add some 170cps for black, but as already black queen and c8 bishop come into the game slower, no need to give more than 10-15cps for better development, so black should be roughly ahead by some 20-30cps in the early mg, probably not sufficient for a win.

Of course, this is all a matter of speculation, there is not a perfect eval and a perfect search, all is just a matter of individual assessment, with occasionally particular assessments being closer to the truth than others, to only fail at other very specific position.

[hgm]

So Komodo's eval is a lot better than yours. No surprise there.

Back-rank orthogonal movers like (limited-range) Rooks indeed have somewhat depressed empirical values, when they start behind a closed rank of Pawns. But this always seems to amount to ~25cP, even in the case of the Wazir. Note that black also has an inactive Rook, so that the difference really only is the inactivity of the Queen-side Rook, and that 50cP for that is a gross over-estimate.

[Lyudmil Tsvetkov]

So Komodo's eval is a lot better than yours. No surprise there.

Back-rank orthogonal movers like (limited-range) Rooks indeed have somewhat depressed empirical values, when they start behind a closed rank of Pawns. But this always seems to amount to ~25cP, even in the case of the Wazir. Note that black also has an inactive Rook, so that the difference really only is the inactivity of the Queen-side Rook, and that 50cP for that is a gross over-estimate.

As stated, my point of view has been confirmed by LTC games, but unfortunately, you seem to read between the lines.