Would you like to react to this message? Create an account in a few clicks or log in to continue.

Chess2uLog in

descriptionRocinante EmptyRocinante

Rocinante Rocina10

by Antonio Torrecillas

descriptionRocinante EmptyRe: Rocinante

Antonio Torrecillas wrote:

"Rocinante 2.0 is available at

Extracted from the readme.txt

Rocinante 2.0 is a cuasi complete rewrite of Rocinante.
There have been some refactoring and changes to optimize the
execution of the program without significantly altering the
underlying algorithms.
The version 1.01 was bitboard and this new release uses mailbox.
Some code has been stolen from Simplex (Zobrist, hashtable, PVS etc).
A new algorithm has been added. MCTS_AB: Monte Carlo Tree Search alpha beta.
Monte Carlo programs are successful in the game of Go, but are notoriously
ineffective in chess. I intend to add this algorithm to verify the veracity
of these claims.
I've added a new evaluation function. And a SeeEval that replaces a quiesce search.
This is configured using UCI protocol options.

2. UCI Options.
id name Rocinante 2.0
id author Antonio Torrecillas
option name cpus type spin default 1 min 0 max 256
option name threads type spin default 1 min 0 max 256
option name klevel type spin default 1 min 0 max 1
option name mcts_ab type check default true
option name probedepth type spin default 1 min -1 max 10

Option cpus and threads control the number of worker threads.

Option mcts_ab control the main algorithm.
true -> use of mcts_ab Monte Carlo Tree Search
false -> use PB* Probability Based B star.

Option klevel control the evaluation function
0 -> UFO material + Piece square tables.
1 -> Material + development + PST + Mobility + ...

Option probedepth
Both MCTS_AB and PB* have an alpha beta algorithm as a replacement
for a static evaluation. Probe Depth control the depth of this alpha beta.
-1 -> no alpha beta is used. Instead a static exchange evaluation is used.
0 -> quiesce evaluation. (MVLA).No checks.
1 -> one ply + quiesce.
2 -> two ply alpha beta + quiesce.
3 -> 3 plys AB+quiesce.
4 and up -> iterative deepening + LMR + nullmove etc.

Unfortunately I can't tell which parameter set is better for each time control. There is a hardcoded limit of 8000 expands.If this limit is reached, consider increasing the depth probe. A minimum of expanded nodes are needed for a correct work of the algorithm. (for PB* 112 expands).

Known bug: Multi procesor is broken in Linux platforms."

descriptionRocinante EmptyRe: Rocinante

Rocinante 2.0 JA (UCI)

Windows/Linux/Android 64/32
A UCI engine based on the interesting & rarely used
algorithm PB* .B* probability based search.

descriptionRocinante EmptyRe: Rocinante

Permissions in this forum:
You cannot reply to topics in this forum