monty v1.0.0

What's New?

• General game playing (with some reasonable success in ataxx).
• Search tree's memory usage now constrained by the Hash UCI option, this means you can now run at any time control without fear!
• Started networks from scratch, meaning monty's data is now fully-original!
• No longer uses Quiescence Search for value, meaning monty is now truly separate from alpha-beta engines!

ELO Gains

Most of the time on this release was spent catching up with the previous version, attempting to compete with external data and qsearch is quite difficult. The current result is that this version's policy network is far superior, whilst the value still lacks somewhat behind.

FRC

Thanks to generating a small amount of FRC data, the gains here are quite large.

Elo   | 167.12 +- 16.17 (95%)
Conf  | 8.0+0.08s Threads=1 Hash=64MB
Games | N: 2002 W: 1341 L: 446 D: 215
Penta | [50, 50, 302, 153, 446]
https://chess.swehosting.se/test/6848/

Standard

Moderate gains, it is unclear how these will scale to VVLTC due to the weaker value network.

Elo   | 45.24 +- 18.69 (95%)
Conf  | 40.0+0.40s Threads=1 Hash=128MB
Games | N: 1004 W: 430 L: 300 D: 274
Penta | [46, 80, 169, 112, 95]
https://chess.swehosting.se/test/6850/

monty v0.1.0

First release of monty, a CPU MCTS engine that plays at around 3k CCRL Blitz.

The MCTS search is still pretty simplistic, so it is not recommended to allow monty to play at VLTC (more than 5 minute games, let's say), as memory usage will increase without bound.

EDIT: I accidentally built the release binary using an old commit, and have now updated it, the only functional change is addition of DFRC support.