Help with transposition table

[eboatwright]

Very good starting point Run a test: disable the transposition table (remove the calls to it, or make sure it never stores anything; I can do this by setting the size to 0 MB) and and run it against the same engine with the TT enabled. The Elo-difference should be somewhere between 130-160 Elo. If so, you could then implement buckets and aging.

Alright. thanks!

[Desperado]

The Elo-difference should be somewhere between 130-160 Elo. If so, you could then implement buckets and aging.

That depends very much on the current level of the engine. You might want to break it down into the following chunks.

1. Move Ordering - retrieving the transposition table move and sorting the movelist but do not prune anything
2. TT pruning - if a transposition is found, pruning the tree
3. Using TT in quiescense search
4. Different search type (pvs, alpha-beta - the handlig might be different)

These are at least the large components that can (and should) be tested separately. If a component does not result in a significant increase in Elo, something is not working properly during sorting and/or pruning. In QS, for example, the benefit night not compensate for the effort.

I guess even for an engine in early stage of devlopment the components should sum up something bewtween 50-100 Elo.

Good Luck, enjoy.

[eboatwright]

3. Using TT in quiescense search

I actually haven't looked into using the TT during the quiescence search, I'll definitely try that out!

[mvanthoor]

I guess even for an engine in early stage of devlopment the components should sum up something bewtween 50-100 Elo.

More, even in a super-basic engine. I actually tested it, in Rustic Alpha 2.

SPRT results

Move ordering on the TT-move adds about 102 elo:

Code: Select all

Score of Rustic Alpha 2 vs Rustic Alpha 1.5: 409 - 197 - 133  [0.643] 739
...      Rustic Alpha 2 playing White: 218 - 84 - 68  [0.681] 370
...      Rustic Alpha 2 playing Black: 191 - 113 - 65  [0.606] 369
...      White vs Black: 331 - 275 - 133  [0.538] 739
Elo difference: 102.5 +/- 23.5, LOS: 100.0 %, DrawRatio: 18.0 %
SPRT: llr 2.95 (100.1%), lbound -2.94, ubound 2.94 - H1 was accepted

TT-cuts (early return from alpha-beta if an evaluation value is found) adds 41 Elo

Code: Select all

Score of Rustic Alpha 1.5 vs Rustic Alpha 1.1: 869 - 646 - 357  [0.560] 1872
...      Rustic Alpha 1.5 playing White: 431 - 321 - 184  [0.559] 936
...      Rustic Alpha 1.5 playing Black: 438 - 325 - 173  [0.560] 936
...      White vs Black: 756 - 759 - 357  [0.499] 1872
Elo difference: 41.6 +/- 14.2, LOS: 100.0 %, DrawRatio: 19.1 %
SPRT: llr 2.95 (100.3%), lbound -2.94, ubound 2.94 - H1 was accepted

Principal Variation Search adds 54 Elo:

Code: Select all

Score of Rustic Alpha 2.2.100 vs Rustic Alpha 2.1.100: 591 - 388 - 318  [0.578] 1297
...      Rustic Alpha 2.2.100 playing White: 334 - 171 - 143  [0.626] 648
...      Rustic Alpha 2.2.100 playing Black: 257 - 217 - 175  [0.531] 649
...      White vs Black: 551 - 428 - 318  [0.547] 1297
Elo difference: 54.8 +/- 16.6, LOS: 100.0 %, DrawRatio: 24.5 %
SPRT: llr 2.95 (100.3%), lbound -2.94, ubound 2.94 - H1 was accepted

Total 197 Elo in self-play. Self-play is always somewhat exaggerated; in a gauntlet, these features add roughly 150 Elo, +/- 10.