Add Late Move Reductions and Principal Variation Search#56
Open
Add Late Move Reductions and Principal Variation Search#56
Conversation
- LMR: Late quiet moves (index >= 3, depth >= 3) are searched with reduced depth using a precomputed log-based reduction table. If the reduced search fails high, re-search at full depth. - PVS: First move searched with full alpha-beta window (PV move). All subsequent moves use null window (-alpha-1, -alpha). If the null window search fails high, re-search with full window. - Together these dramatically reduce the search tree by skipping full-depth searches on moves unlikely to improve the position. Benchmark (depth 4, 48 positions): - Nodes: 4,760,507 → 1,936,234 (−59.3%) - Time: 210.32s → 84.57s (−59.8%)
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 SummaryThis PR adds two standard search optimizations to the alpha-beta engine:
Both techniques are well-established in chess engine development and are implemented correctly following standard patterns. The
Confidence Score: 4/5
|
| Filename | Overview |
|---|---|
| moonfish/engines/alpha_beta.py | Adds LMR (precomputed log-based reduction table, quiet-move filtering) and PVS (null-window + re-search pattern) to negamax. Implementation follows standard chess programming patterns. No logical bugs found; caches in_check to avoid redundant call. |
Flowchart
flowchart TD
A[Start Move Loop] --> B{move_index == 0?}
B -->|Yes| C[Full Window Search\n-beta, -alpha]
B -->|No| D{Is Quiet & Not In Check\n& depth >= 3 & index >= 3?}
D -->|Yes| E[Compute LMR Reduction\nfrom precomputed table]
D -->|No| F[reduction = 0]
E --> G[PVS Null Window Search\ndepth - 1 - reduction\n-alpha-1, -alpha]
F --> G
G --> H{reduction > 0\n& score > alpha?}
H -->|Yes| I[LMR Re-search\nFull Depth, Null Window\n-alpha-1, -alpha]
H -->|No| J{score > alpha\n& score < beta?}
I --> J
J -->|Yes| K[PVS Re-search\nFull Depth, Full Window\n-beta, -alpha]
J -->|No| L[Use Current Score]
K --> L
C --> L
L --> M[Update Alpha/Beta\nCheck Cutoffs]
Last reviewed commit: 2b69491
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
⚡ 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
Two complementary search optimizations that together reduce nodes searched by ~60%:
reduction = floor(log(depth) * log(move_index) / 2). If the reduced search finds a score above alpha, re-search at full depth to verify.(-alpha-1, -alpha). If a null-window search finds a score that improves alpha, re-search with the full window.Tactical moves (captures, promotions, checks) and moves made while in check are never reduced by LMR.
Benchmark (depth 4, 48 positions)
Local Stockfish Benchmark
Settings: 20 games, Stockfish skill 3, 10s/move, no opening book.
Use
/run-stockfish-benchmarkfor CI validation with opening book and longer time control.Test plan
/run-nps-benchmarkfor CI validation/run-stockfish-benchmarkfor strength validation