tdmDefaultsFill: Default values for list `tdm`.

Description

This list controls the tuning and unbiased evaluation phase. When called with tdm = tdmDefaultsFill(), a new list tdm is created and returned. When called with tdm = tdmDefaultsFill(mainFile="my.r", a new list tdm is created and returned, with the element mainFile set to the specified value. When called with tdm = tdmDefaultsFill(tdm), an existing list tdm is filled with further default values.

Usage

tdmDefaultsFill(tdm = NULL, mainFile = NULL)

Arguments

tdm

(optional)

mainFile

(optional) if given, create or overwrite tdm$mainFile with this value

Value

tdm the new / extended list, where additional elements, if they are not yet def'd, are set as:

mainFile

[NULL] if not NULL, source this file from the current dir. It should contain the definition of tdm$mainFunc.

mainFunc

sub(".r","",basename(tdm$mainFile),fixed=TRUE), if tdm$mainFile is set and tdm$mainFunc is NULL, else "mainFunc" This is the name of the function called in tdmStartSpot2 and unbiasedRun

unbiasedFunc

["unbiasedRun"] which function to call for unbiased evaluation

tuneMethod

["spot"] other choices: "cmaes", "bfgs", ..., see tdmDispatchTuner

nExperim

[1]

umode

["RSUB"], one out of [ "RSUB" | "CV" | "TST" | "SP_T" ], see unbiasedRun

timeMode

[1] 1: proc time, 2: system time, 3: elapsed time (columns Time.TST and Time.TRN in envT$theFinals

filenameEnvT

filename where tdmBigLoop will save a small version of environment envT. If NULL, it is set to sub(".conf",".RData",tdm$runList[1]).

theSpotPath

[NA] use SPOT's package version

parallelCPUs

[1] 1: sequential, >1: parallel execution with this many CPUs (package parallel)

parallelFuncs

[NULL] in case tdm$parallelCPUs>1: a string vector with functions which are clusterExport'ed in addition to tdm$mainFunc.

path

[NULL] from where to load and save envT resp. filenameEnvT. If it is NULL, tdm$path is set to the actual working directory at the time when tdmEnvTMakeNew is executed.

runList

[NULL] a list of configuration names .conf

stratified

[NULL] see tdmReadAndSplit

tdmPath

[NULL] from where to source the R sources. If NULL load library TDMR instead.

test2.string

["default cutoff"]

optsVerbosity

[0] the verbosity for the unbiased runs

withParams

[TRUE] list the columns with tuned parameter in final results

nrun

[5] number of runs for unbiased run

U.saveModel

[TRUE] if TRUE, save the last model, which is trained in unbiasedRun, onto filenameEnvT

tstCol

["TST.COL"] opts$TST.COL for unbiased runs (only for umode="TST")

nfold

[10] number of CV-folds for unbiased runs (only for umode="CV")

TST.trnFrac

[NULL] train set fraction (of all train-vali data),OVERWRITES opts$TST.trnFrac if not NULL.

TST.valiFrac

[NULL] validation set fraction (of all train-vali data), OVERWRITES to opts$TST.valiFrac if not NULL.

TST.testFrac

[0.2] test set fraction (of *all* data) for unbiased runs (only for umode="RSUB" or ="SP_T")

CMA.propertyFile

[NULL] (only for CMA-ES Java tuner) see cma_jTuner.

CMA.populationSize

[NULL] (only for CMA-ES Java tuner) see cma_jTuner.

Details

If tdm$mainFunc is missing, but tdm$mainFile exists, then tdmDefaultsFill will set

tdm$mainFunc=sub(".r","",basename(tdm$mainFile),fixed=TRUE)