Fix transposition table and add iterative deepening#54
Open
Fix transposition table and add iterative deepening#54
Conversation
- Remove alpha/beta from TT key (was causing near-zero cache hit rate) - Store entry type (EXACT, LOWER_BOUND, UPPER_BOUND) for proper TT probing with bound-based cutoffs - Add iterative deepening: search depths 1..N, populating TT each iteration so the next iteration has better move ordering - Place TT best move first in move ordering (organize_moves) - Update LazySMP to use new TT key format Benchmark (depth 4, 48 positions): - Nodes: 4,760,507 → 2,834,290 (−40.5%) - Time: 210.32s → 123.86s (−41.1%)
Owner
Author
|
/run-nps-benchmark |
BenchmarksThe following benchmarks are available for this PR:
Post a comment with the command to trigger a benchmark run. |
Greptile SummaryFixed the transposition table key to use only the board hash (removing alpha/beta), added proper TT entry types (EXACT/LOWER/UPPER), implemented iterative deepening, and integrated TT move-first ordering to improve alpha-beta pruning efficiency.
Issues found:
Confidence Score: 2/5
|
| Filename | Overview |
|---|---|
| moonfish/engines/alpha_beta.py | Fixed TT key format and added iterative deepening, but has unreachable code in TT entry type logic and missing draw detection |
| moonfish/move_ordering.py | Added TT move-first ordering, but doesn't validate TT move legality before insertion (hash collision risk) |
| moonfish/engines/lazy_smp.py | Updated to use new TT key format, but bypasses iterative deepening by calling negamax directly at full depth |
Flowchart
flowchart TD
A[search_move called] --> B[Initialize empty TT cache]
B --> C{Iterative Deepening Loop<br/>d=1 to negamax_depth}
C --> D[negamax depth=d]
D --> E{TT key in cache?}
E -->|Yes| F{TT depth >= current depth?}
F -->|Yes| G{Check TT entry type}
G -->|EXACT| H[Return cached score]
G -->|LOWER| I[Update alpha, check cutoff]
G -->|UPPER| J[Update beta, check cutoff]
E -->|No| K[Search position]
F -->|No| K
I --> K
J --> K
K --> L[Organize moves<br/>TT move first]
L --> M{For each move}
M --> N[Recursive negamax]
N --> O{Score >= beta?}
O -->|Yes| P[Store TT_LOWER, return]
O -->|No| Q{Best score found?}
Q --> R{All moves done?}
R -->|No| M
R -->|Yes| S{Determine entry type}
S -->|score <= orig_alpha| T[Store TT_UPPER]
S -->|else| U[Store TT_EXACT]
T --> V[Return to iterative loop]
U --> V
P --> V
H --> V
V --> W{More depths?}
W -->|Yes| C
W -->|No| X[Return best_move]
Last reviewed commit: aada48a
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| # Determine TT entry type based on the score relative to original bounds | ||
| if best_score <= original_alpha: | ||
| tt_type = TT_UPPER # Failed low: score is an upper bound | ||
| elif best_score >= beta: |
There was a problem hiding this comment.
unreachable: if best_score >= beta, would have returned at line 302
Suggested change
| elif best_score >= beta: | |
| elif best_score >= beta: | |
| # This branch is unreachable - beta cutoffs return early at line 302 | |
| tt_type = TT_LOWER # Failed high: score is a lower bound |
Comment on lines
230
to
236
| if board.is_checkmate(): | ||
| cache[cache_key] = (-self.config.checkmate_score, None) | ||
| cache[tt_key] = (-self.config.checkmate_score, None, depth, TT_EXACT) | ||
| return (-self.config.checkmate_score, None) | ||
|
|
||
| if board.is_stalemate(): | ||
| cache[cache_key] = (0, None) | ||
| cache[tt_key] = (0, None, depth, TT_EXACT) | ||
| return (0, None) |
There was a problem hiding this comment.
missing draw detection: should check is_repetition(), is_fifty_moves(), and is_insufficient_material() before checkmate/stalemate (like in quiescence_search:114)
Comment on lines
+39
to
+40
| if tt_move is not None: | ||
| result.insert(0, tt_move) |
There was a problem hiding this comment.
TT move legality not checked: hash collision could insert illegal move. Verify tt_move in board.legal_moves before inserting
Additional Comments (1)
|
⚡ NPS Benchmark Results
Per-position breakdown |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
(alpha, beta), which meant the same position at the same depth would get different cache entries depending on the alpha-beta window. This resulted in near-zero cache hit rates. Now uses just the board position hash as the key.organize_moves(), which dramatically improves alpha-beta cutoff rates.Benchmark (depth 4, 48 positions)
Local Stockfish Benchmark
Settings: 20 games, Stockfish skill 3, st=60, concurrency=4.
Note: This PR was benchmarked with different settings (st=60, concurrency=4) than the other PRs (10s/move, concurrency=1), so results are not directly comparable. The regression may be related to the different test conditions rather than a true strength regression.
Use
/run-stockfish-benchmarkfor CI validation with opening book and longer time control.Test plan
pytest tests/ -k "not lazy_smp and not layer_1 and not layer_2")/run-nps-benchmarkfor CI validation/run-stockfish-benchmarkfor strength validation