Jouni wrote:I noticed, that Critter doesn't use partial 5 piece Gaviota bases. If I remember correct 1.2 used like most other engines. Is it done by purpose, Richard?
I always silently assumed the full set (4 or 5) is installed. Full 5 piece set takes only 7 GB - not much by todays standards. If partial 5 piece worked correctly with a previous Critter version, it was not by design rather by pure luck
Once 6-piece Gaviotas will be available then of course partial set will be more common. I will think out something by then
Jouni wrote:
And there is seems to be something broken. See this mate in 11 position:
6k1/3N4/6K1/7n/8/B7/8/8 w - - 0 1
Analysis by Critter 1.4 64-bit - no tablebases
1.Bd6 Nf4+ 2.Bxf4 Kh8 3.Kf7 Kh7 4.Nf8+ Kh8 5.Be5#
+- (#5) Depth: 44/35 00:00:04 12688kN
Wrong mate.
Thanks for reporting this. There is a bug in the function which evaluates two minors vs one minor endgames. It makes an erroneous assumption that the following position is a checkmate:
Code: Select all
position fen 4n2k/6B1/5NK1/8/8/8/8/8 b - -
eval
eval = -M0
evaluated by endgame knowledge
The corrected version gives
Code: Select all
position fen 4n2k/6B1/5NK1/8/8/8/8/8 b - -
eval
eval = +0.00
evaluated by endgame knowledge
Now the analysis from the original position looks correct:
Code: Select all
27/46 0:11 +M21 1.Bc1 Ng7 2.Nf6+ Kh8 3.Ng4 Nh5 4.Ne5 Kg8
5.Kxh5 Kg7 6.Bg5 Kf8 7.Bd8 Ke8 8.Bf6 Kf8
9.Kg6 Ke8 10.Kf5 Kf8 11.Ke6 Ke8
12.Nf7 Kf8 13.Nd6 Kg8 14.Kf5 (40.109.611) 3568
28/46 0:17 +M11 1.Bc1 Ng7 2.Nf6+ Kh8 3.Bh6 Nf5 4.Bf8 Ne7+
5.Kf7 Nf5 6.Ne4 Kh7 7.Ng5+ Kh8 8.Ba3 Nd6+
9.Kf8 Nc4 10.Be7 Na5 11.Bf6+ (63.510.821) 3596
This fix will be included in a future release.
Jouni wrote:
Analysis by Critter 1.4 64-bit with TBs (everywhere, 4 piece):
1.Kxh5 Kf7 2.Nb6 Ke6 3.Kg4 Ke5 4.Bb4 Kd4 5.Kf3 Kd3 6.Bc5 Kd2 7.Ke4 Kd1 8.Nc4 Ke2 9.Be3 Kf1 10.Kf3 Ke1 11.Bf2+ Kd1 12.Ke3 Kc2 13.Nb6 Kb3 14.Be1 Kc2 15.Ba5 Kb1 16.Kd2 Kb2 17.Bc3+ Kb3 18.Kd3 Ka3 19.Kc2 Ka2 20.Na4 Ka3 21.Nb2 Ka2 22.Bb4 Ka1 23.Nd3 Ka2 24.Nc1+ Ka1 25.Bc3#
+- (#25) Depth: 1/1 00:00:00
Engine should not stop, but continue to find shortest mate I think.
In game play it does not make sense to search when there is a known TB win at ply 1 no matter how long. But in infinite analysis mode, you are right.
Richard