exe Namespace Reference

Functions
DynMat< real_t >	create_transition_matrix ()
DynMat< real_t >	compute_matrix_power (const DynMat< real_t > &mat, uint_t power)
void	print_matrix (const DynMat< real_t > &mat)
real_t	get_reward (real_t prob, uint_t n=10)
void	update_record (std::vector< std::vector< real_t > > &records, uint_t action, real_t r)
uint_t	get_best_arm (const std::vector< std::vector< real_t > > &records)
std::vector< real_t >	get_probs (uint_t n)
DynVec< real_t >	extract_part (const std::vector< std::vector< real_t > > &values)

Variables
const uint_t	N = 10
const auto	N_EXPERIMENTS = 500
const auto	TAU = 0.7
const uint	SEED = 42

Detailed Description

Solve the multi-arm bandit problem using soft-max policy. When using a soft-max policy policy we get a distribution of probabilities over the actions. We select the action with the highest probability. For this example we will solve a 10-armed bandit problem, so N=10.

This example is taken from the book: Reinforcement Learning in Action by Manning Publications.

Function Documentation

compute_matrix_power()

DynMat< real_t > exe::compute_matrix_power	(	const DynMat< real_t > &	mat,
		uint_t	power
	)

create_transition_matrix()

DynMat< real_t > exe::create_transition_matrix

(

)

extract_part()

DynVec< real_t > exe::extract_part

(

const std::vector< std::vector< real_t > > &

values

)

get_best_arm()

uint_t exe::get_best_arm

(

const std::vector< std::vector< real_t > > &

records

)

get_probs()

std::vector< real_t > exe::get_probs

(

uint_t

n

)

get_reward()

real_t exe::get_reward	(	real_t	prob,
		uint_t	n = 10
	)

print_matrix()

void exe::print_matrix

(

const DynMat< real_t > &

mat

)

update_record()

void exe::update_record	(	std::vector< std::vector< real_t > > &	records,
		uint_t	action,
		real_t	r
	)

Variable Documentation

N

const uint_t exe::N = 10

N_EXPERIMENTS

const auto exe::N_EXPERIMENTS = 500

SEED

const uint exe::SEED = 42

TAU

const auto exe::TAU = 0.7