hyperband: Hyperband

In ja-thomas/hyperbandr:

Description

Runs hyperband

Usage

hyperband(problem, max.resources = 81, prop.discard = 3, max.perf = TRUE,
  id, par.set, sample.fun, init.fun, train.fun, performance.fun, ...)

Arguments

problem	[list()] List containing data and other information required for train
max.resources	[integer()] The maximum amount of resource that canbe allocated to a single configuration
prop.discard	[integer()] An input that controls the proportion of configurations discarded in each round of successive halving
max.perf	[logical()] TRUE if to maximize the performance (e.g. accuracy of a neural net), FALSE if to minimize the performance (e.g. find the minimum of the branin function). Default: TRUE
id	[string] An id for each Algorithm object in the bracket object
par.set	The parameter set to sample from
sample.fun	The function to sample from par.set
init.fun	The function to initial a model
train.fun	The function to carry out training
performance.fun	The function to measure the performance
...	Further arguments

Value

List of brackets

Examples

# we need some packages
library("ggplot2")
library("smoof")
library("data.table")

# we choose the 2 dimensional branin function
braninProb = makeBraninFunction()

# the branin function has 3 global minima
opt = data.table(x1 = getGlobalOptimum(braninProb)$param$x1,
  x2 = getGlobalOptimum(braninProb)$param$x2)
param.set = getParamSet(braninProb)


#######################################
## define functions to use hyperband ##
#######################################

# config space
configSpace = makeParamSet(
    makeNumericParam(id = "x1", lower = -5, upper = 10.1))

# sample fun
sample.fun = function(par.set, n.configs, ...) {
  sampleValues(par = par.set, n = n.configs)
}

# init fun
init.fun = function(r, config, problem) {
  x1 = unname(unlist(config))
  x2 = runif(1, 0, 15)
  mod = c(x1, x2)
  return(mod)
}

# train fun
train.fun = function(mod, budget, problem) {
  for(i in seq_len(budget)) {
    mod.new = c(mod[[1]], mod[[2]] + rnorm(1, sd = 3))
    if(performance.fun(mod.new) < performance.fun(mod))
      mod = mod.new
  }
  return(mod)
}

# performance fun
performance.fun = function(model, problem) {
  braninProb(c(model[[1]], model[[2]]))
}

hyperhyper = hyperband(
 problem = braninProb,
 max.resources = 81,
 prop.discard = 3,
 max.perf = FALSE,
 id = "branin",
 par.set = configSpace,
 sample.fun =  sample.fun,
 init.fun = init.fun,
 train.fun = train.fun,
 performance.fun = performance.fun)

# get the best performance of each bracket
lapply(hyperhyper, function(x) x$getPerformances())

ja-thomas/hyperbandr documentation built on May 6, 2019, 8:33 p.m.