Hyperion - new UCI chess engine
Author: twhite1457
Hyperion 0.2.0-beta what's new?
Hyperion 0.2.0-beta: Puzzle Elo and Hand-Crafted Evaluation
This second beta release of Hyperion introduces major new features for engine evaluation and search, alongside a massive performance uplift. The focus of this update is on providing better tools for measuring strength and experimenting with different search strategies.
The engine continues to be in beta, with the core MCTS and UCI functionality being the primary focus for testing.
What's New
Puzzle-Based Elo Rating System
A new puzzle evaluation mode has been added to gauge the engine's tactical ability. This system allows Hyperion to solve a series of chess puzzles to determine an Elo rating.
Adaptive Testing: The puzzle difficulty is adaptively adjusted based on the engine's performance, providing a more accurate strength assessment.
Glicko-2 Rating: It uses the Glicko-2 rating system to calculate the engine's Elo, including a rating deviation (margin of error) to show the confidence in the rating.
Hand-Crafted Evaluation MCTS (MCE-MCTS)
A new search mode has been implemented that uses a basic, hand-crafted evaluation function to guide the Monte Carlo Tree Search. This provides a tangible improvement in playing strength over a "pure" MCTS. Users can now test and play against both the pure MCTS and this new evaluation-guided search.
Massive Performance Gains
The core engine's move generation and perft calculations have been significantly optimized.
Perft test speeds have increased dramatically, now reaching nearly 44 million nodes per second (4.39454e+07 nps). This improvement makes search deeper and more efficient.
Comments
Post a Comment