Detailed Description

internal methods for epsilon greedy bandit selection

Author: Gregor Hendel

#include "blockmemshell/memory.h"
#include "scip/def.h"
#include "scip/type_bandit.h"
#include "scip/type_retcode.h"
#include "scip/type_scip.h"

Go to the source code of this file.

Functions
SCIP_RETCODE	SCIPincludeBanditvtableEpsgreedy (SCIP *scip)

	SCIP_DECL_BANDITFREE (SCIPbanditFreeEpsgreedy)

	SCIP_DECL_BANDITSELECT (SCIPbanditSelectEpsgreedy)

	SCIP_DECL_BANDITUPDATE (SCIPbanditUpdateEpsgreedy)

	SCIP_DECL_BANDITRESET (SCIPbanditResetEpsgreedy)

SCIP_RETCODE	SCIPbanditCreateEpsgreedy (BMS_BLKMEM blkmem, BMS_BUFMEM bufmem, SCIP_BANDITVTABLE vtable, SCIP_BANDIT epsgreedy, SCIP_Real priorities, SCIP_Real eps, SCIP_Bool preferrecent, SCIP_Real decayfactor, int avglim, int nactions, unsigned int initseed)

Function Documentation

◆ SCIPincludeBanditvtableEpsgreedy()

SCIP_RETCODE SCIPincludeBanditvtableEpsgreedy ( SCIP * scip )

creates the epsilon greedy bandit algorithm includes it in SCIP

Parameters

scip	SCIP data structure

Definition at line 331 of file bandit_epsgreedy.c.

References BANDIT_NAME, SCIP_CALL, SCIP_OKAY, and SCIPincludeBanditvtable().

Referenced by SCIPincludeCorePlugins().

◆ SCIP_DECL_BANDITFREE()

SCIP_DECL_BANDITFREE ( SCIPbanditFreeEpsgreedy )

callback to free bandit specific data structures

Definition at line 59 of file bandit_epsgreedy.c.

References BMSfreeBlockMemory, BMSfreeBlockMemoryArray, NULL, SCIP_OKAY, SCIPbanditGetData(), SCIPbanditGetNActions(), and SCIPbanditSetData().

◆ SCIP_DECL_BANDITSELECT()

SCIP_DECL_BANDITSELECT ( SCIPbanditSelectEpsgreedy )

selection callback for bandit algorithm

Definition at line 82 of file bandit_epsgreedy.c.

References EPSGREEDY_SMALL, NULL, SCIP_OKAY, SCIP_Real, SCIPbanditGetData(), SCIPbanditGetNActions(), SCIPbanditGetRandnumgen(), SCIPrandomGetInt(), SCIPrandomGetReal(), and sqrt().

◆ SCIP_DECL_BANDITUPDATE()

SCIP_DECL_BANDITUPDATE ( SCIPbanditUpdateEpsgreedy )

update callback for bandit algorithm

Definition at line 145 of file bandit_epsgreedy.c.

References NULL, SCIP_OKAY, SCIP_Real, and SCIPbanditGetData().

◆ SCIP_DECL_BANDITRESET()

SCIP_DECL_BANDITRESET ( SCIPbanditResetEpsgreedy )

reset callback for bandit algorithm

Definition at line 181 of file bandit_epsgreedy.c.

References BMSclearMemoryArray, EPSGREEDY_SMALL, NULL, SCIP_OKAY, SCIP_Real, SCIPbanditGetData(), SCIPbanditGetNActions(), SCIPbanditGetRandnumgen(), SCIPrandomGetReal(), and w.

◆ SCIPbanditCreateEpsgreedy()

SCIP_RETCODE SCIPbanditCreateEpsgreedy	(	BMS_BLKMEM *	blkmem,
		BMS_BUFMEM *	bufmem,
		SCIP_BANDITVTABLE *	vtable,
		SCIP_BANDIT **	epsgreedy,
		SCIP_Real *	priorities,
		SCIP_Real	eps,
		SCIP_Bool	preferrecent,
		SCIP_Real	decayfactor,
		int	avglim,
		int	nactions,
		unsigned int	initseed
	)

internal method to create and reset epsilon greedy bandit algorithm

Parameters

blkmem	block memory
bufmem	buffer memory
vtable	virtual function table with epsilon greedy callbacks
epsgreedy	pointer to store the epsilon greedy bandit algorithm
priorities	nonnegative priorities for each action, or NULL if not needed
eps	parameter to increase probability for exploration between all actions
preferrecent	should the weights be updated in an exponentially decaying way?
decayfactor	the factor to reduce the weight of older observations if exponential decay is enabled
avglim	nonnegative limit on observation number before the exponential decay starts, only relevant if exponential decay is enabled
nactions	the positive number of possible actions
initseed	initial random seed

Definition at line 233 of file bandit_epsgreedy.c.

References BMSallocBlockMemory, BMSallocBlockMemoryArray, eps, NULL, SCIP_ALLOC, SCIP_CALL, SCIP_OKAY, and SCIPbanditCreate().

Referenced by SCIPcreateBanditEpsgreedy().

SCIP