Fetching latest commit…
NRPA
This is a parallel implementation of the NRPA algorithm (Nested Rollout Policy Allocation).
Build
cd src
make
Run a test (same game)
cd src
./same
Advanced testing
Run all tests
make test
Show the average score curve (requires gnuplot)
make run420
make show420
Debug
Changing
CXXFLAGS=-O3 -g -DNDEBUG -lpthread -std=c++11
into
CXXFLAGS=-O0 -g -lpthread -std=c++11
In src/Makefile will run in debug mode. It is recommanded to run in debug mode before.
How to use NRPA to solve a new game
-
Define a Board class and a Move class following as in leftmove.cpp
-
Include "nrpa.hpp"
-
In the main function, call:
Nrpa<Board, Move, 5, MaxPlayoutLength, MaxLegalMoves>::test(Options::parse(argc, argv));
where:
- 'Board' is your board class
- 'Move' is your move class
- '5' is the maximum Nrpa depth + 1
- 'MaxPlayoutLength' is an upper bound on the number of moves in a single game
- 'MaxLegalMoves' is an upper bound on the number of distinct legal moves
-
Compile your program
-
run it.
Many examples are available in src/test/
Usage
Usage: ./same [options]
where option is any of these:
--num-run=NUM, -r NUM
Number of test runs (default: 4).
--num-level=NUM, -l NUM
Nrpa depth (default: 4).
--num-thread=NUM, -x NUM
Number of threads (0 = hardware capacity, default: 0).
--timeout=NUM, -t NUM
Timeout in sec. for a single run (0 = no timeout, default: 0).
--iter-stats, -s
Enable iteration based statistics (default: no).
--timer-stats, -S
Enable timer based statistics (default: no).
--tag=STRING, -T STRING
Set a tag name for this run, it will be appened to statistics file name (default: None).
--parallel-level=NUM, -p NUM
Go parallel call at level N (default: 1).
--parallel-strat=NUM, -P NUM
Use parallelization strategy number N (default: 1).
--thread-stats, -q
Enable thread statistics (default: 0).
--help, -h
This help.
Authors
- Benjamin Negrevergne
- Tristan Cazenave
License
GPL3