startMnistRun: Start hyperparameter optimization runs with spot based on...

In SPOTMisc: Misc Extensions for the 'SPOT' Package

Start hyperparameter optimization runs with spot based on MNIST data

Description

Runs to compare deep learning models. Note: Number of epochs is limited: model <- "dl"; cfg <- getModelConf(model = model); cfg$upper[6] <- 5

Usage

startMnistRun(
  runNr = "000",
  SPOTVersion = "2.11.4",
  SPOTMiscVersion = "1.19.6",
  encoding = "tensor",
  network = "cnn",
  timebudget = 60,
  data.seed = 1,
  prop = 2/3,
  batch_size = 32,
  tuner.seed = 1,
  returnValue = "validationLoss",
  initSizeFactor = 1,
  spotModel = buildKriging,
  spotOptim = optimDE,
  lower = NULL,
  upper = NULL,
  noise = TRUE,
  OCBA = FALSE,
  OCBABudget = 0,
  multiStart = 2,
  multFun = 200,
  handleNAsMethod = handleNAsMean,
  imputeCriteriaFuns = list(is.infinite, is.na, is.nan),
  krigingTarget = "ei",
  krigingUseLambda = TRUE,
  krigingReinterpolate = TRUE,
  defaultAsStartingPoint = TRUE,
  plots = FALSE,
  Rinit = 1,
  replicates = 1,
  resDummy = FALSE,
  verbosity = 0
)

Arguments

runNr	character, specifies the run number. Default: "000"
SPOTVersion	smallest package version number
SPOTMiscVersion	smallest package version number
encoding	encoding: "oneHot" od "tensor". Default: "tensor"
network	network: "dl" odr "cnn". Default: "cnn"
timebudget	time budget Default: 3600 (secs)
data.seed	1
prop	proportion 2 / 3
batch_size	batch size (for dl) 32
tuner.seed	seed for SPOT 1
returnValue	"validationLoss"
initSizeFactor	multiplier for the initial design size 2
spotModel	buildKriging
spotOptim	optimDE
lower	NULL
upper	NULL
noise	TRUE
OCBA	TRUE
OCBABudget	3
multiStart	2
multFun	200
handleNAsMethod	handleNAsMean
imputeCriteriaFuns	list(is.infinite, is.na, is.nan)
krigingTarget	"ei"
krigingUseLambda	TRUE
krigingReinterpolate	FALSE
defaultAsStartingPoint	FALSE
plots	FALSE
Rinit	2
replicates	2
resDummy	FALSE
verbosity	0

Examples

### These examples require an activated Python environment as described in
### Bartz-Beielstein, T., Rehbach, F., Sen, A., and Zaefferer, M.:
### Surrogate Model Based Hyperparameter Tuning for Deep Learning with SPOT,
### June 2021. http://arxiv.org/abs/2105.14625.
PYTHON_RETICULATE <- FALSE
if(PYTHON_RETICULATE){
library("dplyr")
library("farff")
library("GGally")
library("keras")
library("tensorflow")
library("Metrics")
library("mlr")
library("reticulate")
library("rpart")
library("rpart.plot")
library("SPOT")
library("SPOTMisc")
library("tfdatasets")
library("rsample")
startMnistRun(timebudget=60, initSizeFactor = 1, verbosity = 1)
startMnistRun(timebudget=60, encoding="tensor", network="cnn")
}

SPOTMisc documentation built on Sept. 5, 2022, 5:06 p.m.