privateRF: Private random forests algorithm

In insilico/privateEC: Privacy Preserving Evaporative Cooling Feature Selection and Classification with Relief-F and Random Forests

Description

Random Forest Thresholdout, which is TO with the feature selection and classifier replaced with Random Forest.

Usage

privateRF(
  train.ds = NULL,
  holdout.ds = NULL,
  validation.ds = NULL,
  label = "class",
  is.simulated = TRUE,
  rf.importance.measure = "MeanDecreaseGini",
  rf.ntree = 500,
  rf.mtry = NULL,
  pec.file = NULL,
  update.freq = 50,
  threshold = 4/sqrt(nrow(train.ds)),
  tolerance = 1/sqrt(nrow(train.ds)),
  signal.names = NULL,
  save.file = NULL,
  verbose = FALSE
)

Arguments

train.ds	A data frame with training data and outcome labels
holdout.ds	A data frame with holdout data and outcome labels
validation.ds	A data frame with validation data and outcome labels
label	A character vector of the outcome variable column name
is.simulated	Is the data simulated (or real?)
rf.importance.measure	A character vector for the random forest importance measure
rf.ntree	An integer the number of trees in the random forest
rf.mtry	An integer the number of variables sampled at each random forest node split
pec.file	A character vector filename of privateEC results
update.freq	A integer for the number of steps before update
threshold	A numeric, default 4 / sqrt(n) suggested in the thresholdout’s supplementary material (Dwork, et al.,2015)
tolerance	A numeric, default 1 / sqrt(n) suggested in the thresholdout’s supplementary material (Dwork, et al.,2015)
signal.names	A character vector of signal names in simulated data
save.file	A character vector for results filename or NULL to skip
verbose	A flag indicating whether verbose output be sent to stdout

Value

A list containing:

algo.acc

data frame of results, a row for each update

ggplot.data

melted results data frame for plotting

correct

number of variables detected correctly in each data set

elapsed

total elapsed time

See Also

Other classification: epistasisRank(), getImportanceScores(), originalThresholdout(), privateEC(), standardRF(), xgboostRF()

Examples

num.samples <- 100
num.variables <- 100
pct.signals <- 0.1
label <- "class"
temp.pec.file <- tempfile(pattern = "pEc_temp", tmpdir = tempdir())
sim.data <- createSimulation(num.variables = num.variables,
                             num.samples = num.samples,
                             pct.signals = pct.signals,
                             label = label,
                             sim.type = "mainEffect",
                             pct.train = 1 / 3,
                             pct.holdout = 1 / 3,
                             pct.validation = 1 / 3,
                             verbose = FALSE)
pec.results <- privateEC(train.ds = sim.data$train,
                         holdout.ds = sim.data$holdout,
                         validation.ds = sim.data$validation,
                         label = sim.data$label,
                         is.simulated = TRUE,
                         signal.names = sim.data$signal.names,
                         save.file = temp.pec.file,
                         verbose = FALSE)
prf.results <- privateRF(train.ds = sim.data$train,
                         holdout.ds = sim.data$holdout,
                         validation.ds = sim.data$validation,
                         label = sim.data$label,
                         is.simulated = TRUE,
                         signal.names = sim.data$signal.names,
                         pec.file = temp.pec.file,
                         verbose = FALSE)

insilico/privateEC documentation built on May 22, 2020, 5:12 p.m.