Author: Ulrich Tuerke
Date: 06:10:31 02/11/03
Go up one level in this thread
On February 10, 2003 at 16:21:23, Sune Fischer wrote:
>I'm fooling around with some root node aspiration search,
>Looks like it is a tad unstable, I wonder if that is normal.
>
>It produces different preferred moves at different plies.
>E.g. normally it prefers e4 at ply 11, but using a narrow window of 0.1 pawn it
>prefers e4 already at ply 9. Sure the hashtable is going to make a difference
>here, I realize that, but still a change on the root move is not an
>insignificant effect.
>
>The narrower the window, the more instability. Is this to be expected?
I don't think that this is unusual.
The source of these search anomalies are dependencies of the search on the
current values of alpha and beta. These could be:
lazy eval (as Robert mentioned),
null move pruning,
futility (and other) forward cuts,
extensions which are triggered by comparing current score to alpha/beta
...
Theoretically, the search is only guaranteed to be stable, if you eliminate
these dependencies. (I have decided to live with them.)
Uli
>
>the code:
>
>window = param::pawn_unit/10;
>walpha = previous_score-window;
>wbeta = previous_score+window;
>score = -b.Search(1,ONEPLY*(maxdepth-1),-wbeta,-walpha,0);
>if (score <= walpha) {
> score = -b.Search(1,ONEPLY*(maxdepth-1),-wbeta,-alpha,0);
> if (score >= wbeta)
> score = -b.Search(1,ONEPLY*(maxdepth-1),-beta,-alpha,0);
>} else if (score >= wbeta) {
> score = -b.Search(1,ONEPLY*(maxdepth-1),-beta,-walpha,0);
> if (score <= walpha)
> score = -b.Search(1,ONEPLY*(maxdepth-1),-beta,-alpha,0);
>}
>
>
>results of different window widths:
>
>window=pawn_unit:
>
>1 -1 0 1 1.a3
>1 14 1 5 1.d3
>1 23 1 14 1.d4
>1 105 1 21 1.Nc3
>2 -103 1 52 1.Nc3 Nf3
>2 0 2 96 1.Nf3 Nf6
>3 103 2 280 1.Nf3 Nf6 2.Nc3
>4 -23 2 507 1.Nf3 Nc3
>4 23 3 1535 1.Nc3 Nf6 2.Nf3
>5 23 3 1711 1.Nc3 Nf6 2.Nf3
>6 0 4 2781 1.Nc3 Nf6 2.Nf3 Nc6 3.e4 e5
>7 16 5 7890 1.Nc3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Bf4
>8 6 6 14544 1.Nc3 Nf6 2.Nf3 Nc6 3.e4 d5 4.e5 Ng4
>9 15 33 197084 1.Nc3 d5 2.Nf3 Nc6
>10 8 66 439047 1.Nc3 Nf6 2.e4 e5 3.Nf3 Nc6
>11 12 234 1641245 1.Nc3 Nf6 2.Nf3 Nc6 3.d4 d5 4.e3 Bf5 5.Bd3 Bxd3 6.Qxd3
>11 15 597 4247723 1.e4 e5 2.Nf3 Nc6
>12 12 992 7037705 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bb5 exd4
>13 14 2475 17533424 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bb5 Nd4 5.O-O Bc5 6.Nxe5 Nxb5
>7.Nxb5 Nxe4
>14 14 3768 26153111 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bb5 Nd4 5.O-O Bc5 6.Nxe5 Nxb5
>7.Nxb5 Nxe4
>
>window=pawn_unit/10:
>
>1 -1 0 1 1.a3
>1 14 1 5 1.d3
>1 23 1 14 1.d4
>1 105 1 21 1.Nc3
>2 -103 2 52 1.Nc3 Nf3
>2 0 2 96 1.Nf3 Nf6
>3 103 2 272 1.Nf3 Nf6 2.Nc3
>4 -23 2 489 1.Nf3 Nc3
>4 23 3 1517 1.Nc3 Nf6 2.Nf3
>5 23 3 1693 1.Nc3 Nf6 2.Nf3
>6 -16 4 3765 1.Nc3 Nf3
>6 16 4 4367 1.Nf3 Nf6 2.Nc3
>7 16 5 8324 1.Nf3 Nf6 2.Nc3
>8 6 7 15958 1.Nf3 Nf6 2.Nc3 Nc6 3.e4 d5 4.e5 Ng4 5.Nxe4
>9 15 21 112696 1.Nf3 Nf6 2.Nc3 d5 3.d4 Nc6 4.Bf4 Bf5 5.e3
>9 17 41 249717 1.e4 Nc6 2.Nc3 Nf6 3.Nf3 d5 4.exd5 Nxd5 5.Bd3
>10 8 110 729380 1.e4 Nf6 2.Nc3 e5 3.Nf3 Nc6 4.Nc3
>11 15 392 2758226 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.Bxc6 dxc6 5.Nc3 Bd6 6.O-O
>12 12 740 5225754 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bb5 Bd6 5.O-O a6 6.Bxc6 dxc6
>13 14 2689 19137433 1.e4 e5 2.Nc3 Nf6 3.Nf3
>14 14 3521 25053272 1.e4 e5 2.Nc3 Nf6 3.Nf3
>
>window=pawn_unit/100: (pawn_unit=1000)
>
>
>1 -1 0 1 1.a3
>1 14 0 5 1.d3
>1 23 0 14 1.d4
>1 105 1 21 1.Nc3
>2 -206 1 49 1.Nc3 Nf3
>2 0 1 93 1.Nf3 Nf6
>3 103 1 271 1.Nf3 Nf6 2.Nc3
>4 -23 2 553 1.Nf3 Nc3
>4 23 2 1629 1.Nc3 Nf6 2.Nf3
>5 23 2 1803 1.Nc3 Nf6 2.Nf3
>6 -16 3 3382 1.Nc3 Nf3
>6 16 4 3829 1.Nf3 Nf6 2.Nc3
>7 16 4 7702 1.Nf3 Nf6 2.Nc3
>8 -15 7 26519 1.Nf3 Nc3
>8 -11 10 37593 1.a3 Nf6 2.Nc3 Nc6 3.Nf3 d5 4.d4 Bf5 5.Nxe4
>8 17 15 76685 1.e4 Nf6 2.Nc3 Nc6 3.Nf3
>9 17 31 185188 1.e4 d5 2.Nc3 Nf6
>10 -24 86 596412 1.e4 exd5
>10 0 106 742042 1.d4 d5 2.Nc3 Nf6 3.Nf3 Nc6 4.Bf4 Bf5 5.Nb5 Rc8 6.Nxe5
>10 8 161 1149018 1.Nc3 Nf6 2.e4 e5 3.Nf3 Nc6
>11 12 347 2482318 1.Nc3 Nf6 2.Nf3 Nc6 3.d4 d5 4.e3 Bf5 5.Bd3 Bxd3 6.Qxd3
>11 15 658 4712297 1.e4 e5 2.Nf3 Nc6
>12 -14 1006 7183908 1.e4 Nc3
>12 14 1222 8702636 1.Nc3 Nf6 2.Nf3 Nc6 3.e4 e5
>13 14 1970 14002541 1.Nc3 Nf6 2.Nf3 Nc6 3.e4 e5
>14 14 3647 25603089 1.Nc3 Nf6 2.Nf3 Nc6 3.e4 e5
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.