stoSOO: R6 Class Representing a function optimizer by StoSOO...

In KosukeHamazaki/myBreedSimulatR: R Breeding Simulator for K.Hamazaki

R6 Class Representing a function optimizer by StoSOO (Stochastic Simultaneous Optimistic Optimization)

Description

stoSOO object store specific information on global optimization for stochastic functions given a finite number of budget (StoSOO).

Public fields

parameter

[vector] Vector with length defining the dimensionality of the optimization problem. Providing actual values of par is not necessary (NAs are just fine).

paramLen

[numeric] Number of parameters

optimizeFunc

[function] Scalar function to be optimized, with first argument to be optimized over

lowerBound

[numeric] Vectors of lower bounds on the variables

upperBound

[numeric] Vectors of upper bounds on the variables

nIterOptimization

[numeric] Number of function evaluations allocated to optimization

nMaxEvalPerNode

[numeric] Maximum number of evaluations per leaf

maxDepth

[numeric] Maximum depth of the tree

nChildrenPerExpansion

[numeric] Number of children per expansion

confidenceParam

[numeric] Confidence parameter (see Valko et al., 2013)

maximize

[logical] If TRUE, performs maximization

optimizeType

[character] Either 'deterministic' for optimizing a deterministic function or 'stochastic' for a stochastic one

returnOptimalNodes

[numeric] When (how many iterations) to return (or save) the optimal nodes when optimizing hyper parameter by StoSOO

saveTreeNameBase

[character] Base name of the tree to be saved

whenToSaveTrees

[numeric] When (how many iterations) to save the tree in StoSOO

withCheck

[logical] Check arguments for 'node', 'layer', and 'tree' class or not

verbose

[logical] Display information

widthBase

[numeric] Base of width of the estimates of rewards

funcScale

[numeric] Scale for function to be optimized. If 'maximize = TRUE', 'funcScale = 1', and else 'funcScale = -1'.

maximizeFunc

[function] Function to be maximized given parameters scaled from 0 to 1.

currentTree

[tree class] 'tree' class object for the current status

optimalNodes

[list] List of optimal nodes ('node' class object) corresponding to 'returnOptimalNodes'

optimalNodeFinal

[node class] Optimal node ('node' class object) for the final tree

optimalParameter

[numeric] Optimal parameter estimated by StooSOO given a finite number of evaluations

optimalValue

[numeric] Optimal value estimated by StooSOO given a finite number of evaluations

Methods

Public methods

• stoSOO$new()

• stoSOO$startOptimization()

• stoSOO$print()

• stoSOO$clone()

---

Method new()

Create a new stoSOO object.

Usage

stoSOO$new(
  parameter,
  optimizeFunc,
  ...,
  lowerBound = NULL,
  upperBound = NULL,
  nIterOptimization = NULL,
  nMaxEvalPerNode = NULL,
  maxDepth = NULL,
  nChildrenPerExpansion = NULL,
  confidenceParam = NULL,
  maximize = NULL,
  optimizeType = NULL,
  returnOptimalNodes = NULL,
  saveTreeNameBase = NULL,
  whenToSaveTrees = NA,
  withCheck = FALSE,
  verbose = TRUE
)

Arguments

parameter

[vector] Vector with length defining the dimensionality of the optimization problem. Providing actual values of par is not necessary (NAs are just fine).

optimizeFunc

[function] Scalar function to be optimized, with first argument to be optimized over

...

[logical/numeric/character/etc...] Optional additional arguments to 'optimizeFunc'

lowerBound

[numeric] Vectors of lower bounds on the variables

upperBound

[numeric] Vectors of upper bounds on the variables

nIterOptimization

[numeric] Number of function evaluations allocated to optimization

nMaxEvalPerNode

[numeric] Maximum number of evaluations per leaf

maxDepth

[numeric] Maximum depth of the tree

nChildrenPerExpansion

[numeric] Number of children per expansion

confidenceParam

[numeric] Confidence parameter (see Valko et al., 2013)

maximize

[logical] If TRUE, performs maximization

optimizeType

[character] Either 'deterministic' for optimizing a deterministic function or 'stochastic' for a stochastic one

returnOptimalNodes

[numeric] When (how many iterations) to return (or save) the optimal nodes when optimizing hyper parameter by StoSOO

saveTreeNameBase

[character] Base name of the tree to be saved

whenToSaveTrees

[numeric] When (how many iterations) to save the tree in StoSOO

withCheck

[logical] Check arguments for 'node', 'layer', and 'tree' class or not

verbose

[logical] Display information

---

Method startOptimization()

start global optimization of stochastic function by StoSOO

Usage

stoSOO$startOptimization()

---

Method print()

Display information about the object

Usage

stoSOO$print()

---

Method clone()

The objects of this class are cloneable with this method.

Usage

stoSOO$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

References

R. Munos (2011), Optimistic optimization of deterministic functions without the knowledge of its smoothness, NIPS, 783-791.

M. Valko, A. Carpentier and R. Munos (2013), Stochastic Simultaneous Optimistic Optimization, ICML, 19-27 http://hal.inria.fr/hal-00789606. Matlab code: https://team.inria.fr/sequel/software/StoSOO.

P. Preux, R. Munos, M. Valko (2014), Bandits attack function optimization, IEEE Congress on Evolutionary Computation (CEC), 2245-2252.

KosukeHamazaki/myBreedSimulatR documentation built on Aug. 31, 2024, 3:55 p.m.