Houdini 3 reducing the depth feature

[Don]

Houdini 3 finds THE score and PV.

The technical idea behind this is the difference in pruning and extensions between PV nodes and non-PV nodes.
If you find a fail-high for a non-PV node at depth 30, it's very likely that when it becomes a PV node the fail-high can be found at reduced depth.

Houdini 3 demonstrates how remarkably effective this idea is.

I'm not sure I understand the idea. Do you do this at the root or recursively in PV nodes? So if you are on iteration 30 you suddenly reset to iteration 30-n on a fail-hi?

Don

[Uri Blass]

Houdini 3 finds THE score and PV.

The technical idea behind this is the difference in pruning and extensions between PV nodes and non-PV nodes.
If you find a fail-high for a non-PV node at depth 30, it's very likely that when it becomes a PV node the fail-high can be found at reduced depth.

Houdini 3 demonstrates how remarkably effective this idea is.

I'm not sure I understand the idea. Do you do this at the root or recursively in PV nodes? So if you are on iteration 30 you suddenly reset to iteration 30-n on a fail-hi?

Don

When I look at the analysis of houdini3 it seems that he does not do it on every fail high.

For reduction that is always by 3 plies based on my experience you need the following conditions:
1)Houdini changes its mind to a different move that earlier it reduced.
2)In the research with bigger window houdini failed high again.

Here is an example from analysis of one of the nolot positions
You can see that Bxh7+ failed high at depth 15 and that it again fail high
so houdini reduced the depth from 15 to 12.

I do not post the rest of the analysis but later houdini got a score for Bxh7+ also at depth 13,14,15,16...


[d]r3rbk1/ppq2ppp/2b1pB2/8/6Q1/1P1B3P/P1P2PP1/R2R2K1 w - - 0 24

15/36 00:02 9,365,253 4,396,000 +0.20 Rd1-d2 e6-e5 Bd3-f5 e5-e4 Ra1-d1

g7-g6 Bf5-d7 e4-e3 f2xe3 Bf8-h6 Bd7xe8 Bh6xe3+ Kg1-h1 Be3xd2 Be8xc6 Qc7xc6 Rd1xd2 Qc6xf6

Qg4-e4 Kg8-g7 Rd2-d7 Ra8-d8 Rd7xd8 Qf6xd8 Qe4xb7 Qd8-d1+ Kh1-h2 Qd1-d6+ g2-g3 Qd6-d2+ Qb7-

g2
15/53+ 00:02 11,990,392 4,517,000 +0.27 Bd3xh7+
15/62+ 00:02 12,903,969 4,534,000 +0.42 Bd3xh7+
other best move - reduce depth to 12
12/62 00:03 12,958,878 4,531,000 +0.56 Bd3xh7+ Kg8xh7 Qg4-h5+ Kh7-g8 Rd1-

d4 g7xf6 Rd4-g4+ Bf8-g7 Qh5-h6 Kg8-f8 Rg4xg7 Re8-c8 Qh6xf6 Bc6-e4 Rg7-g4 Be4xc2 Rg4-c4 Qc7

-d8 Qf6xd8+ Rc8xd8 Rc4xc2 Ra8-c8 Ra1-c1 Rc8xc2 Rc1xc2 Rd8-d7 h3-h4 Kf8-g7

[gerold]

Let me see: Houdini 450-500 Elo stronger than Crafty. Advantage Robert.

Are you sure Houdini Doesn't have a bit of Crafty in it.

[Don]

Let me see: Houdini 450-500 Elo stronger than Crafty. Advantage Robert.

Are you sure Houdini Doesn't have a bit of Crafty in it.

You don't want to go there ....

[michiguel]

Let me see: Houdini 450-500 Elo stronger than Crafty. Advantage Robert.

Are you sure Houdini Doesn't have a bit of Crafty in it.

[MODERATION]
Easy guys...

Miguel

[Uri Blass]

It works! I'm using a similar method in the new Vitruvius (less sophisticated than the one implemented by Robert) and results are surprising. For example, in this position:

FEN: r3b1k1/1rq1bppp/2n1p3/ppPpP3/2pP1P2/P1P1BB2/3Q2PP/1R2NRK1 w - - 0 1

Vitruvius 1.16 (in "Deep-Analysis" mode) finds, surprisingly quickly, the best move: 1.f5!
This without going too deep... I believe that this move is very difficult to find, for any engine ... Except, of course, Vitruvius 1.16 in "Deep-Analysis" mode and Houdini-3 in Tactical Mode.

Another example:

FEN: r1bq1rk1/ppp3b1/3p1nn1/P1PPp1Bp/1P2P3/1QNN1Pp1/4B1P1/R4RK1 b - - 0 1

Vitruvius 1.16 "Deep-Analysis" finds the best move (1.Bh6!) in only 4 minuts, on my computer. Depth: 21 plyes. At the same time Stockfish is at depth=27 plyes... And has not yet found the solution!

I do not own Houdini-3 and then I could not do the tests, but I think between the various engines is the only one who can find, such as Vitruvius 1.16, the solution in reasonable time...

P.S. I'm sorry, but I don't know how you post pictures.

Roberto.

houdini3 tactical does not see f5 in the first diagram(note that I did not check that f5 is the best move)

[d]r3b1k1/1rq1bppp/2n1p3/ppPpP3/2pP1P2/P1P1BB2/3Q2PP/1R2NRK1 w
- - 0 1

Houdini_3_x64:
4/12 00:00 974 30,000 +0.57 Ne1-c2 Ra8-d8 Be3-f2 Be8-d7
5/12 00:00 1,634 49,000 +0.68 Ne1-c2 Ra8-d8 Be3-f2 Be8-d7 Nc2-e3
6/21 00:00 5,174 53,000 +0.74 Ne1-c2 b5-b4 a3xb4 a5xb4 Nc2xb4 Nc6xb4 Rb1xb4 Be8-c6 Rb4xb7 Qc7xb7
7/21 00:00 8,255 71,000 +0.63 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 f4-f5 Ra8-a2 f5xe6 f7xe6
8/21 00:00 10,184 85,000 +0.63 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 f4-f5 Ra8-a2 f5xe6 f7xe6
9/21 00:00 15,715 126,000 +0.64 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Rb1-b2 Be8-d7 Be3-f2 h7-h6
10/22- 00:00 45,788 334,000 +0.56 Ne1-c2 Qc7-d7
10/25 00:00 77,212 518,000 +0.50 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Be3-f2 Bg6-f7 Nc2-e3 b5-b4
11/25 00:00 89,855 572,000 +0.50 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Rf1-e1 Bg6-f7 Be3-f2 Ra8-d8 Nc2-e3
12/25 00:00 158,832 882,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Rf1-e1 Bg6-e8 Be3-f2 Be8-f7 Nc2-e3 b5-b4 c3xb4 a5xb4 a3xb4 Rb7-b5
13/29- 00:00 234,915 1,145,000 +0.47 Ne1-c2 Qc7-d7
13/29 00:00 405,026 1,639,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Ra8-b8 Be3-f2 Be8-f7 Rf1-e1 Rb8-f8 Nc2-e3 Rb7-b8 Qd2-e2
14/29+ 00:00 500,046 1,845,000 +0.55 Ne1-c2
14/29 00:00 593,462 2,025,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Be3-f2 Ra8-b8 Rf1-e1 Rb8-e8 Rb1-b2 Re8-f8 Re1-b1 Rf8-b8 Nc2-e3 b5-b4 a3xb4 a5xb4
15/29 00:00 1,127,480 2,804,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Be3-f2 Ra8-b8 Rf1-e1 Rb8-e8 Rb1-b2 Re8-f8 Re1-b1 Rf8-b8 Nc2-e3 b5-b4 a3xb4 a5xb4
16/30- 00:00 2,114,526 3,524,000 +0.47 Ne1-c2 Qc7-d7
16/37 00:01 8,447,570 4,698,000 +0.59 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Be7-h4 Be3-f2 Bh4xf2+ Qd2xf2 Qd7-e8 Qf2-e2 Qe8-e7 Nc2-e3
17/38- 00:02 9,462,301 4,709,000 +0.51 Ne1-c2 Qc7-d7
17/38 00:02 12,561,887 4,868,000 +0.51 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Qe8-d7 Be3-f2 g7-g5 f4xg5 Be7xg5 Nc2-e3 b5-b4 a3xb4 a5xb4
18/38 00:03 15,420,876 4,987,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
18/59+ 00:08 45,958,840 5,269,000 +0.59 Bf3-g4
18/59 00:09 48,041,611 5,271,000 +0.45 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Rb1-b2 Nc6-a5 Rf1-a1 Na5-b3 Ra1xa8 Qb8xa8 Qd2-c3 Qa8-a4 Bg4-f3 Be8-c6 Be3-f2 Rb7-a7 Rb2-b1 Qa4-a2 Rb1-e1 Ra7-a8
18/59 00:09 48,043,338 5,267,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
19/59 00:10 52,510,154 5,268,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
20/59 00:12 66,379,258 5,316,000 +0.45 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e3 Bh5xf3 Qf2xf3 Qe8-f7 Qf3-f2 b5-b4 a3xb4 a5xb4
21/59 00:22 119,792,088 5,432,000 +0.44 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Qf2-d2 b5-b4 c3xb4 a5xb4 a3xb4 Rb8-a8 b4-b5
21/59+ 00:32 178,862,932 5,506,000 +0.53 Bf3-g4
21/59 00:34 191,598,201 5,512,000 +0.48 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Bf3-d1 Be8-d7 Be3-c1 Bf8-g7 Rf1-f2 Ra2-a7 Bc1-d2 Ra7-a2 Rb1-b2 Ra2-a7 Bd1-f3 Rb7-b5
22/59+ 00:55 307,984,038 5,552,000 +0.56 Bf3-g4
22/59 00:59 327,167,706 5,555,000 +0.46 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Rb7-b5 Bf3-d1 Rb5-b7 Be3-c1 Be7-f8 Bc1-a3 Nc6-e7 Rb1-b2 Ra2xb2 Qc3xb2 Ne7-f5 g2-g4 Nf5-h4 Qb2-c3 Rb7-a7 Ba3-b2 Ra7-a2
23/59+ 01:32 514,863,678 5,561,000 +0.54 Bf3-g4
23/61 02:23 783,384,082 5,475,000 +0.53 Bf3-g4 g7-g6 Ne1-c2 Be7-f8 Bg4-f3 Nc6-e7 Be3-f2 Bf8-g7 Rf1-e1 Be8-c6 Nc2-e3 b5-b4 a3xb4 a5xb4 Rb1xb4 Rb7xb4 c3xb4 Ra8-a3 Bf3-e2 Qc7-a7 Bf2-h4 Ra3-a2 Qd2-c3 Ra2-a3 Qc3-c1 Ra3-b3 Qc1-d2
24/61 03:42 1,220,124,332 5,482,000 +0.59 Bf3-g4 g7-g6 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-d1 h7-h5 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Rd1-f1 Rb7-b5 h2-h3 Qb8-b7 Bh5-f3 Bg7-f8 Rb1-a1 Ra2xa1 Rf1xa1 Nc6xb4 Nc2xb4 Rb5xb4
25/62 06:45 2,239,247,812 5,517,000 +0.59 Bf3-g4 g7-g6 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Qd2-c3 Be7-f8 Bg4-f3 Ra8-a2 Rf1-e1 h7-h5 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Re1-f1 Rb7-b5 h2-h3 Qb8-b7 Bh5-f3 Bg7-f8 Rb1-a1 Ra2xa1 Rf1xa1
26/65- 15:28 5,115,404,863 5,508,000 +0.51 Bf3-g4 b5-b4
26/65 22:20 7,429,551,864 5,544,000 +0.59 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-e1 h7-h5 Re1-c1 Bf8-h6 Rc1-d1 Bh6-f8 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Rd1-f1 Rb7-b5 Rf1-f3 Nc6xe5 d4xe5 Bg7xe5 Nc2-d4 Rb5xc5 g2-g3 Rc5-a5 b4xa5 Qb8xb1+ Rf3-f1
27/65 28:30 9,517,870,527 5,565,000 +0.54 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-d1 Be8-d7 Rb1-b2 Ra2-a4 Rd1-b1 Nc6-a7 Nc2-a3 Na7-b5 Na3xb5 Bd7xb5 Bf3-d1 Ra4-a6 Rb2-f2 Bb5-c6 Be3-c1 Rb7-a7 b4-b5 Bc6xb5 Rf2-b2 Ra7-b7

[Don]

The problem with this is that Komodo wants to play f5 right away, changes it's mind and comes back to it once in a while. So it's not really a good position for this sort of thing. I have not analyzed it but if it wins big and doesn't look like a bad move it's not a good position for testing this particularly idea.

It works! I'm using a similar method in the new Vitruvius (less sophisticated than the one implemented by Robert) and results are surprising. For example, in this position:

FEN: r3b1k1/1rq1bppp/2n1p3/ppPpP3/2pP1P2/P1P1BB2/3Q2PP/1R2NRK1 w - - 0 1

Vitruvius 1.16 (in "Deep-Analysis" mode) finds, surprisingly quickly, the best move: 1.f5!
This without going too deep... I believe that this move is very difficult to find, for any engine ... Except, of course, Vitruvius 1.16 in "Deep-Analysis" mode and Houdini-3 in Tactical Mode.

Another example:

FEN: r1bq1rk1/ppp3b1/3p1nn1/P1PPp1Bp/1P2P3/1QNN1Pp1/4B1P1/R4RK1 b - - 0 1

Vitruvius 1.16 "Deep-Analysis" finds the best move (1.Bh6!) in only 4 minuts, on my computer. Depth: 21 plyes. At the same time Stockfish is at depth=27 plyes... And has not yet found the solution!

I do not own Houdini-3 and then I could not do the tests, but I think between the various engines is the only one who can find, such as Vitruvius 1.16, the solution in reasonable time...

P.S. I'm sorry, but I don't know how you post pictures.

Roberto.

houdini3 tactical does not see f5 in the first diagram(note that I did not check that f5 is the best move)

[d]r3b1k1/1rq1bppp/2n1p3/ppPpP3/2pP1P2/P1P1BB2/3Q2PP/1R2NRK1 w
- - 0 1

Houdini_3_x64:
4/12 00:00 974 30,000 +0.57 Ne1-c2 Ra8-d8 Be3-f2 Be8-d7
5/12 00:00 1,634 49,000 +0.68 Ne1-c2 Ra8-d8 Be3-f2 Be8-d7 Nc2-e3
6/21 00:00 5,174 53,000 +0.74 Ne1-c2 b5-b4 a3xb4 a5xb4 Nc2xb4 Nc6xb4 Rb1xb4 Be8-c6 Rb4xb7 Qc7xb7
7/21 00:00 8,255 71,000 +0.63 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 f4-f5 Ra8-a2 f5xe6 f7xe6
8/21 00:00 10,184 85,000 +0.63 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 f4-f5 Ra8-a2 f5xe6 f7xe6
9/21 00:00 15,715 126,000 +0.64 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Rb1-b2 Be8-d7 Be3-f2 h7-h6
10/22- 00:00 45,788 334,000 +0.56 Ne1-c2 Qc7-d7
10/25 00:00 77,212 518,000 +0.50 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Be3-f2 Bg6-f7 Nc2-e3 b5-b4
11/25 00:00 89,855 572,000 +0.50 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Rf1-e1 Bg6-f7 Be3-f2 Ra8-d8 Nc2-e3
12/25 00:00 158,832 882,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-g6 Rf1-e1 Bg6-e8 Be3-f2 Be8-f7 Nc2-e3 b5-b4 c3xb4 a5xb4 a3xb4 Rb7-b5
13/29- 00:00 234,915 1,145,000 +0.47 Ne1-c2 Qc7-d7
13/29 00:00 405,026 1,639,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Ra8-b8 Be3-f2 Be8-f7 Rf1-e1 Rb8-f8 Nc2-e3 Rb7-b8 Qd2-e2
14/29+ 00:00 500,046 1,845,000 +0.55 Ne1-c2
14/29 00:00 593,462 2,025,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Be3-f2 Ra8-b8 Rf1-e1 Rb8-e8 Rb1-b2 Re8-f8 Re1-b1 Rf8-b8 Nc2-e3 b5-b4 a3xb4 a5xb4
15/29 00:00 1,127,480 2,804,000 +0.55 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Be3-f2 Ra8-b8 Rf1-e1 Rb8-e8 Rb1-b2 Re8-f8 Re1-b1 Rf8-b8 Nc2-e3 b5-b4 a3xb4 a5xb4
16/30- 00:00 2,114,526 3,524,000 +0.47 Ne1-c2 Qc7-d7
16/37 00:01 8,447,570 4,698,000 +0.59 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Be7-h4 Be3-f2 Bh4xf2+ Qd2xf2 Qd7-e8 Qf2-e2 Qe8-e7 Nc2-e3
17/38- 00:02 9,462,301 4,709,000 +0.51 Ne1-c2 Qc7-d7
17/38 00:02 12,561,887 4,868,000 +0.51 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Qe8-d7 Be3-f2 g7-g5 f4xg5 Be7xg5 Nc2-e3 b5-b4 a3xb4 a5xb4
18/38 00:03 15,420,876 4,987,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
18/59+ 00:08 45,958,840 5,269,000 +0.59 Bf3-g4
18/59 00:09 48,041,611 5,271,000 +0.45 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Rb1-b2 Nc6-a5 Rf1-a1 Na5-b3 Ra1xa8 Qb8xa8 Qd2-c3 Qa8-a4 Bg4-f3 Be8-c6 Be3-f2 Rb7-a7 Rb2-b1 Qa4-a2 Rb1-e1 Ra7-a8
18/59 00:09 48,043,338 5,267,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
19/59 00:10 52,510,154 5,268,000 +0.47 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Rb1-d1 b5-b4 a3xb4 a5xb4
20/59 00:12 66,379,258 5,316,000 +0.45 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e3 Bh5xf3 Qf2xf3 Qe8-f7 Qf3-f2 b5-b4 a3xb4 a5xb4
21/59 00:22 119,792,088 5,432,000 +0.44 Ne1-c2 Qc7-d7 Bf3-g4 f7-f5 Bg4-f3 Be8-f7 Rb1-b2 Ra8-b8 Rf1-b1 Qd7-e8 Qd2-e2 Be7-h4 Be3-f2 Bh4xf2+ Qe2xf2 Bf7-h5 Nc2-e1 Bh5-f7 Qf2-d2 b5-b4 c3xb4 a5xb4 a3xb4 Rb8-a8 b4-b5
21/59+ 00:32 178,862,932 5,506,000 +0.53 Bf3-g4
21/59 00:34 191,598,201 5,512,000 +0.48 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Bf3-d1 Be8-d7 Be3-c1 Bf8-g7 Rf1-f2 Ra2-a7 Bc1-d2 Ra7-a2 Rb1-b2 Ra2-a7 Bd1-f3 Rb7-b5
22/59+ 00:55 307,984,038 5,552,000 +0.56 Bf3-g4
22/59 00:59 327,167,706 5,555,000 +0.46 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Rb7-b5 Bf3-d1 Rb5-b7 Be3-c1 Be7-f8 Bc1-a3 Nc6-e7 Rb1-b2 Ra2xb2 Qc3xb2 Ne7-f5 g2-g4 Nf5-h4 Qb2-c3 Rb7-a7 Ba3-b2 Ra7-a2
23/59+ 01:32 514,863,678 5,561,000 +0.54 Bf3-g4
23/61 02:23 783,384,082 5,475,000 +0.53 Bf3-g4 g7-g6 Ne1-c2 Be7-f8 Bg4-f3 Nc6-e7 Be3-f2 Bf8-g7 Rf1-e1 Be8-c6 Nc2-e3 b5-b4 a3xb4 a5xb4 Rb1xb4 Rb7xb4 c3xb4 Ra8-a3 Bf3-e2 Qc7-a7 Bf2-h4 Ra3-a2 Qd2-c3 Ra2-a3 Qc3-c1 Ra3-b3 Qc1-d2
24/61 03:42 1,220,124,332 5,482,000 +0.59 Bf3-g4 g7-g6 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-d1 h7-h5 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Rd1-f1 Rb7-b5 h2-h3 Qb8-b7 Bh5-f3 Bg7-f8 Rb1-a1 Ra2xa1 Rf1xa1 Nc6xb4 Nc2xb4 Rb5xb4
25/62 06:45 2,239,247,812 5,517,000 +0.59 Bf3-g4 g7-g6 Ne1-c2 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Qd2-c3 Be7-f8 Bg4-f3 Ra8-a2 Rf1-e1 h7-h5 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Re1-f1 Rb7-b5 h2-h3 Qb8-b7 Bh5-f3 Bg7-f8 Rb1-a1 Ra2xa1 Rf1xa1
26/65- 15:28 5,115,404,863 5,508,000 +0.51 Bf3-g4 b5-b4
26/65 22:20 7,429,551,864 5,544,000 +0.59 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-e1 h7-h5 Re1-c1 Bf8-h6 Rc1-d1 Bh6-f8 f4-f5 g6xf5 Bf3xh5 Bf8-g7 Rd1-f1 Rb7-b5 Rf1-f3 Nc6xe5 d4xe5 Bg7xe5 Nc2-d4 Rb5xc5 g2-g3 Rc5-a5 b4xa5 Qb8xb1+ Rf3-f1
27/65 28:30 9,517,870,527 5,565,000 +0.54 Bf3-g4 b5-b4 a3xb4 a5xb4 c3xb4 Qc7-b8 Ne1-c2 g7-g6 Qd2-c3 Ra8-a2 Bg4-f3 Be7-f8 Rf1-d1 Be8-d7 Rb1-b2 Ra2-a4 Rd1-b1 Nc6-a7 Nc2-a3 Na7-b5 Na3xb5 Bd7xb5 Bf3-d1 Ra4-a6 Rb2-f2 Bb5-c6 Be3-c1 Rb7-a7 b4-b5 Bc6xb5 Rf2-b2 Ra7-b7

[bob]

Let me see: Houdini 450-500 Elo stronger than Crafty. Advantage Robert.

And that has WHAT to do with the current discussion???

[bob]

How? A simple example. At depth 30, where the fail high occurs, you store TT entries for plies 1, 3, 5, ..., 29 that say "score >= beta". For plies 2, 4, ..., 30, you store TT entries that say "score <= alpha".

When you back up the depth a bit (I do not know how far you back it up, you can supply that number if you want, so I will use 6 for this discussion, to have a number to work with), when you search this particular move, you will see "> beta" but with a big draft. At depth=24, that big draft will swamp the basic search depth and the TT will cause you to fail high, but the search won't be able to relax beta and get a real score at only depth 24. I don't see how you are going to get a true score until you get back to depth=30.

Simply don't do hash cut-offs in pv-nodes and it will work just fine.

What is this obsession with "let's not do it right so things will work right?" I see absolutely ZERO justification to say "don't do hash cutoffs ANYWHERE."

Unless you say you extend the PV itself for some reason, but not the other moves.

That is exactly what is happening in most modern engines. There is no pruning in PV nodes and reductions are more conservative. PV search to nominal depth N actually tends to be much deeper than a non-pv search to the same nominal depth N.

I understand the idea, although I do not buy "all in" that it is correct. Why would one do things so differently in the PV, that you make it very hard for a non-PV move to fail high in the first place??? I don't personally assume that this is the way things have to be done, it's just the way some do it at the moment. From a theoretical perspective, it seems wrong, however.

[DocNZ]

Another example:

FEN: r1bq1rk1/ppp3b1/3p1nn1/P1PPp1Bp/1P2P3/1QNN1Pp1/4B1P1/R4RK1 b - - 0 1

Vitruvius 1.16 "Deep-Analysis" finds the best move (1.Bh6!) in only 4 minuts, on my computer. Depth: 21 plyes. At the same time Stockfish is at depth=27 plyes... And has not yet found the solution!


Roberto.

When will Vitruvius 1.16 be released? Its the only engine I use (v1.11H).

Paul T.