SPOTMisc

1.19.52

• OpenML dependency removed. Since version == 1.19.52 the package OpenML is not imported any more, because it is not maintained on CRAN. If OpenML is available again, you can uncomment the lines in getDataCensus.R and rebuild the package. The census data set is included as dataCensusFull

1.19.50

• URL: https://keras.rstudio.com/ (moved to https://tensorflow.rstudio.com/)

1.19.48

• prepareComparisonPlot accepts extended parameter list:
  • defaultModelList and tunedModelList

1.19.46

• kerasConf\(returnObject == "model" returns y instead of only the model, i.e., y\)model

1.19.44

• BUGFIX: dropoutfac value in getModelConf() corrected: [0.5; 1] -> [0; 0.5]

1.19.42

• BUGFIX:
  • prepareProgressPlot() uses information about initial number of replicates (Rinit)

1.19.40

• NEW:
  • getModelConf: number of units modified: 0;5 (1;32)

1.19.38

• NEW:
  • further tests tfdatasets

1.19.36

• NEW:
  • funKerasGeneric: embedding enhanced

1.19.32

• BUGFIX: “mace” -> “race” in Census Data

1.19.30

• BUGFIX: *funKerasGeneric: embedding corrected

1.19.28

• BUGFIX:
  • xgboost added to suggests in DESCRIPTION

1.19.26

• NEWS:
  • new function trans_odd_round()
  • evalKerasMnist() checks size (simple check)
  • new transformations for model “cnn” in getModelConf(): kernel uses trans_odd_round()

1.19.24

• resDl100 data

1.19.22

• NEWS:
  • getKerasConf() moved to getKerasConf.R
  • getKerasConf(): new argument tfDevice = “/cpu:0”
  • getMnistData(), evalKerasGeneric() and funKerasGeneric() require the argument kerasConf (getKerasConf() is not used as the default)
  • https addresses in documentations use
  • funKerasModular.R renamed to funKerasGeneric.R

1.19.20

• NEWS:
  • getModelConf() arguments modified. Function accepts a list, e.g., from getKerasConf()

1.19.18:

• NEWS:
  • documentation

1.19.16:

• NEWS
  • dataCensus: data frame
  • getDataCensus() uses SPOTMisc data frame “dataCensus” if getDataCensus() fails

1.19.14

• NEWS
  • spotKeras()

1.19.12

• NEWS:
  • pooling modified

1.19.10

• NEWS:
  • kerasActivation()

1.19.8

• NEWS
  • hyperparameter handling extended: mapX2Flags() uses new argument “model”. Default: “dl”

1.19.6

• NEWS:
  • limited number of training epochs in startMnistRun.R

1.19.4

• NEWS:
  • MNIST updated
  • kConf -> kerasConf
  • startMnistRun()

1.19.2

• BUGFIX: tryCatch -> try in evalKerasGeneric()

1.19.0

• NEWS:
  • evalParamCensus() accepts new parameter xbest. Default: “xBestOcba”

1.18.14

• NEWS:
  • cleanup. funKerasGeneric.R removed

1.18.12

• NEWS:
  • getMlConfig() does not call getDataCensus(), but accepts data as an argument.
  • get_objf() renamed to getObjf()

1.18.10

• NEWS:
  • test predDlCensus

1.18.8

• BUGFIX:
  • predMlCensus transforms x values before calculation
• NEWS:
  • getDataCensus: new argument “cache.only” with default value “FALSE”
  • documentation, cleanup

1.18.6

• BUGFIX: prop corrected in evalParamCensus()

1.18.4

• NEW: evalParamCensus() added back

1.18.2

• NEWS: startKerasRun()

1.18.0

• NEWS: funKerasModular

1.17.14

• NEWS:
  • documentation

1.17.12

• NEWS: handleNA -> handleNAs

1.17.10

• NEWS:
  • kerasConf: naDummy removed, use spot::handleNAs instead

1.17.8

• NEWS: startCensusRun implemented as a script, not a function

1.17.6

• NEWS: kerasEvalPrediction() uses testScore instead of Metrics

1.17.4

• NEWS: predictDL rewritten

1.17.2

• NEWS:
  • CPU/GPU handling

1.17.0

• NEWS:
  • tf updated

1.16.4

• NEW:
  • with(tf$device(“/cpu:0”), {}) is not used anymore, instead the follwoing mechanism is used: library(keras) use_session_with_seed(42, disable_gpu = FALSE, disable_parallel_cpu = FALSE)

1.16.2

• NEWS:
  • target set in getDataCensus

1.16.0

• NEWS:
  • Generic keras functions

1.15.24

• BUGFIX:
  • parallelPlot fixed
• NEWS:
  • spotStats
  • keras functions use “with(tf$device(”/cpu:0”), {})” to force CPU usage

1.15.22

• BUGFIX:
  • OCBAbudget -> OCBABudget (typo)

1.15.20

• NEWS:
• plots updated

1.15.18

• NEWS:
  • OCBA

1.15.16

• BUGFIX:
  • commented code

1.15.14

• BUGFIX:
  • getParamConf: missing task.type added

1.15.12

• NEWS:
  • evalParamCensus() determines nFeatures
  • cache.dir set to “oml.cache”

1.5.10

• BUGFIX: example evalParamCensus fixed

1.15.8

• BUGFIX: evalParamCensus(): x is converted to matrix

1.15.6

• NEWS: Documentation.

1.15.4

• NEWS: startCensusRun()

1.15.2

• NEWS:
  • Improved experimental designs and evaluations for keras

1.15.0

• NEWS:
  • prepare optimizer select. This is hyperparameter x11, which can encoded as type=factor.

1.14.2

• BUGFIX:
  • predDlCensus uses test data

1.14.0

• BUGFIX:
  • getMlrResample() gets task information, i.e.: getMlrResample(task,..)

1.13.6

• NEWS:
  • further test (data census)

1.13.4

• NEWS:
  • further test (data census)

1.13.2

• NEW: additional tests

1.13.0

• NEW: funKerasCensus.R: tests

1.12.6

• NEW: getModelConf() can handle dl, too funKerasCensus accepts target argumet (default = NULL)

1.12.4

• NEW: trans_mult_2 function

1.12.2

• BUGFIX:
  • corrected minbucket default value (1/3)

1.12.0

• NEWS:
  • getModelConf() includes default values (defaults)

1.11.18

• NEWS: cleanup and documentation:
  • test_ds_census is not used

1.11.16

• NEWS:
  • deep learning on census data uses embedding layer

1.11.14

• BUGFIX: fixing continued

1.11.12

• BUGFIX: still incomplete fix

1.11.10

• BUGFIX: target specification

1.11.8

• BUGFIX: prop handling in getCensusTrainValTestData()

1.11.6

• NEW: default param (only begin)

1.11.4

• NEW: predfun() cleanup

1.11.2

• NEW:
  • predMlCensus rewritten (old version: predMlCensusDeprecated)

1.11.0

• NEW:
  • mlr pipeline rewritten

1.10.4

• BUGFIX
  • filter -> dplyr::filter

1.10.2

• BUGFIX
  • predMlCensus() uses testData for task definition

1.10.0

• NEWS:
  • predMlCensus() rewritten
  • new function: makeLearnerFromHyperparameters()
  • plot_pareto removed
  • package emoa not in DESCRIPTION

1.9.6

• BUGFIX in predMlCensus(): trueY col nums fixed
• getCensusTrainValTestData requires target argument
• target income_class replaced by age

1.9.4

• BUGFIX in predMlCensus(): duplicate task generation

1.9.2

• NEWS:
  • predlMlCensus
  • scorePredictions

1.9.0

• NEWS:
  • mlr cleanup and restructure

1.8.12

• BUGFIX:
  • Usage of “name” in plotSensitivity() fixed

1.8.10

• NEWS:
  • New plot functions (based on ggplot2)

1.8.8

• NEWS: test funKeras: test for adam

1.8.6

• NEWS: getCensusTrainValTestData(): prop can be a 2-dim vector

1.8.4

• NEWS:
  • accuracy uses as.integer(pred >0.5)

1.8.2

• NEWS:
  • Extended paprameter conf fpr Census DNNs

1.8.0

• NEWS:
  • Census data for Keras

1.7.6

• BUGFIX:
  • funKerasCensus: several bug fixes

1.7.4

• NEWS:
  • funKerasCensus()
  • dataKeras.R with census data

1.7.2

• NEWS:
  • More documentation
  • get_task_census() is based on getMlrTask() and getDataCensus()

1.7.0

• NEWS
  • hyperparameter tuning functions added
  • get_task_census(): get OpenML KDD Census data set
  • getExplan()

1.6.6

• NEWS:
  • resKerasMnist07 data added
  • resKerasMnist11 prepared (multi start spot)

1.6.4

• NEWS:
  • resKerasMnist03 updated
  • experiments 07-10 prepared

1.6.2

• NEWS
  • resKerasMnist02 updated

1.6.0

• BUGFIX:
  • apply() in funKerasMnist uses “byrow=TRUE”
• NEWS:
  • Extended logging facilities

1.5.20

• NEWS:
  • vignette examples updated, resKerasMnist02 uses transformX values
  • resKerasMnist02 updated
  • default resKerasMnist02 calculation uses transfromX

1.5.18

• NEWS: bachtsize parameter in funKerasMinst() is not mapped to 2^batchsize, because this is handled by SPOT via transferX()

1.5.14 and 16

• NEWS:
  • resKerasMnist02 updated

1.5.12

• NEWS:
  • getMnistData()

1.5.10

• CHANGES:
  • Updated results for kerasMnist02/03

1.5.8

• BUGFIX
  • NA handling fixed

1.5.6

• BUGFIX
  • correct treatment of negative accuracy values

1.5.4

• BUGFIX
  • dim of MOO retrun values fixed

1.5.2

• NEWS
  • keras mock-up functions (resKerasDummy) for quick testing

1.5.0

• NEWS
  • SIR modeling and Markov analyis removed

1.4.4

• BUGFIX
  • SIR modelling: tests/examples skipped

1.4.2

• CHANGES:
  • funKerasMnist and funKerasTransferLearning return y (instead of y and a model list)

1.4.0

• NEW:
  • Cleanup

1.3.12

• NEW:
  • Data sets resKerasTransferLearning* updated
  • Data set resKerasTransferLearning06 prepared

1.3.10

• NEW:
  • Additional examples

1.3.8

• NEW:
  • neg val accuracy

1.3.6

• NEW:
  • handle NaN and Inf: mapped to 1234567890

1.3.4

• BUGFIX:
  • beta* -> beta_*

1.3.2

• NEW:
  • kerasOptimizer.R

1.3.0

• NEW:
  • Perform and document keras examples
  • Basic vignette
  • Additional keras data sets

1.2.30

• NEW:
  • kerasHelper.R

1.2.28

• NEW:
  • function funKerasTransferLearning

1.2.26

• BUGFIX:
  • correct return value in funKerasMnist “res$y”

1.2.22

• NEW:
  • test for funKeras*

1.2.20

• NEW
  • funKeras* functions extended

1.2.18

• BUGFIX:
  • learning_rate replaces lr
• Roxygen 7.1.2

1.2.16

• NEW:
  • spotPlot

1.2.14

• CHANGES:
  • Documentation improved

1.2.12

• BUGFIX:
  • funKerasMnist returns the correct test loss
  • funKerasMnist accepts further keras configuration parameters such as verbose etc. as arguments: funKerasMnist <- function (x, kConf = getKerasConf()).

1.2.10

• CHANGES:
  • funKerasMnist returns a list. score can be accessed via $score

1.2.8

• NEW:
  • funKerasMnist accepts additional parameters:
    • verbose
    • callbacks
      
    • validation_split
    • validation_data
      
    • shuffle Logical

1.2.6

• NEW:
  • funKerasMnist (requires a working python enviroment via reticulate)

1.2.4

• BUGFIX:
  • Namespaces in Imports field not imported from:‘MASS’ ‘RColorBrewer’ ‘ggplot2’ All declared Imports should be used.