Parameters for GoUctGlobalSearchState. More...
#include <GoUctGlobalSearch.h>
Public Member Functions | |
| GoUctGlobalSearchStateParam () | |
Public Attributes | |
| bool | m_mercyRule |
| Use the mercy rule. | |
| bool | m_territoryStatistics |
| Compute probabilities of territory in terminal positions. | |
| SgUctValue | m_lengthModification |
| Modify game result by the length of the simulation. | |
| SgUctValue | m_scoreModification |
| Modify game result by score. | |
Parameters for GoUctGlobalSearchState.
Definition at line 35 of file GoUctGlobalSearch.h.
| GoUctGlobalSearchStateParam::GoUctGlobalSearchStateParam | ( | ) |
Definition at line 12 of file GoUctGlobalSearch.cpp.
Modify game result by the length of the simulation.
This modifies the win/loss result (1/0) by the length of the game counted as number of moves from the root position of the search. This can help to prefer shorter games (if it is a win) or longer games (if it is a loss). It can also have a positive effect on the playing strength, because the variance of results is larger for longer games. The modification is added for losses and subtracted for wins. The value is the length of times the value of the modification parameter. The maximum modification is 0.5. The default value of the parameter is 0.
Definition at line 55 of file GoUctGlobalSearch.h.
Referenced by GoUctCommands::CmdParamGlobalSearch(), GoUctGlobalSearchState< POLICY >::EvaluateBoard(), and GoUctGlobalSearch< POLICY, FACTORY >::SetDefaultParameters().
Use the mercy rule.
Count games early as win or loss, if stone difference on board exceeds a threshold of 30% of the total number of points on board.
Definition at line 40 of file GoUctGlobalSearch.h.
Referenced by GoUctGlobalSearchState< POLICY >::CheckMercyRule(), GoUctCommands::CmdParamGlobalSearch(), GoUctGlobalSearchState< POLICY >::EvaluateBoard(), and GoUctGlobalSearchState< POLICY >::GeneratePlayoutMove().
Modify game result by score.
This modifies the win/loss result (1/0) by the score of the end position. The modification is added for losses and subtracted for wins. The modification value is the score divided by the maximum score, which can be reached on the board, times the value of the score modification parameter. This helps to play moves to maximize the score even if the game is already clearly lost or won. Otherwise all moves look equal in clearly won or lost positions. It can also reduce the typical game length and could even have a positive effect on the playing strength. The modification can be disabled by setting the parameter to zero. The default value is 0.02.
Definition at line 68 of file GoUctGlobalSearch.h.
Referenced by GoUctCommands::CmdParamGlobalSearch(), and GoUctGlobalSearchState< POLICY >::EvaluateBoard().
Compute probabilities of territory in terminal positions.
Definition at line 43 of file GoUctGlobalSearch.h.
Referenced by GoUctCommands::CmdParamGlobalSearch(), GoUctCommands::DoFinalStatusSearch(), GoUctGlobalSearchState< POLICY >::EvaluateBoard(), and GoUctGlobalSearch< POLICY, FACTORY >::OnStartSearch().