MaskJointDensity (version 1.0)

MaskJointDensity-package: Masking, unmasking and restoring confidential data.

Description

This package has 3 key functionalities. It is able to mask confidential data using multiplicative noise. It is able to unmask this data while still preserving confidentiality. It is able to calculate the numerical joint density function of the original data from the unmasked data, as well as obtaining a sample from the marginal density functions of the unmasked data. The final results are a reasonable approximation to the original data for the purposes of analysis.

Details

Package:	MaskJointDensity
Type:	Package
Version:	1
Date:	2018-05-05
License:	GPL-2

The Data Provider obtains confidential data, then masks it and writes it. The End-User unmasks this data then can obtain the joint density function of the unmasked variables and samples from the marginal density of them in order to mimic having the confidential data for the purpose of statistical analysis.

References

Yan-Xia Lin, Luke Mazur, Rathin Sarathy, Krishnamurty Muralidhar Statistical Information Recovery from Multivariate Noise-Multiplied Data, a Computational Approach Transactions on Data Privacy 11:1 (2018) 23 - 45 http://www.tdp.cat/issues16/abs.a271a17.php

Examples

# NOT RUN {
# Data Provider
set.seed(100)
data(FPI2002Data)
OriginalVar01 <- log(FPI2002Data$FPIflow2)
OriginalVar02 <- log(FPI2002Data$gdp_o)  
OriginalVar03 <- log(FPI2002Data$gdp_d) 
OriginalVar04 <- log(FPI2002Data$distw) 

testN <- nrow(FPI2002Data)

noise<-rmulti(n=testN, mean=c(80, 100), sd=c(5,3), p=c(0.6, 0.4))
xmask<-mask(vectorToBeMasked=OriginalVar01, noisefile=file.path(tempdir(),"noise1.bin"),
noise=noise, lowerBoundAsGivenByProvider=min(OriginalVar01),
upperBoundAsGivenByProvider=max(OriginalVar01))
write(xmask$ystar, file.path(tempdir(),"xstar1.dat"))

noise<-rmulti(n=testN, mean=c(60, 90), sd=c(5,3), p=c(0.4, 0.6))
xmask<-mask(vectorToBeMasked=OriginalVar02, noisefile=file.path(tempdir(),"noise2.bin"),
noise=noise, lowerBoundAsGivenByProvider=min(OriginalVar02),
upperBoundAsGivenByProvider=max(OriginalVar02))
write(xmask$ystar, file.path(tempdir(),"xstar2.dat"))

noise<-rmulti(n=testN, mean=c(40, 110), sd=c(5,3), p=c(0.6, 0.4))
xmask<-mask(vectorToBeMasked=OriginalVar03, noisefile=file.path(tempdir(),"noise3.bin"),
noise=noise, lowerBoundAsGivenByProvider=min(OriginalVar03),
upperBoundAsGivenByProvider=max(OriginalVar03))
write(xmask$ystar, file.path(tempdir(),"xstar3.dat"))

noise<-rmulti(n=testN, mean=c(70, 100), sd=c(5,3), p=c(0.4, 0.6))
xmask<-mask(vectorToBeMasked=OriginalVar04, noisefile=file.path(tempdir(),"noise4.bin"),
noise=noise, lowerBoundAsGivenByProvider=min(OriginalVar04),
upperBoundAsGivenByProvider=max(OriginalVar04))
write(xmask$ystar, file.path(tempdir(),"xstar4.dat"))
# }
# NOT RUN {
# End-User
# }
# NOT RUN {
xstar1 <- scan(file.path(tempdir(),"xstar1.dat"))
Unmasked01 <- unmask(maskedVectorToBeUnmasked=xstar1,
noisefile=file.path(tempdir(),"noise1.bin"))

xstar2 <- scan(file.path(tempdir(),"xstar2.dat"))
Unmasked02 <- unmask(maskedVectorToBeUnmasked=xstar2,
noisefile=file.path(tempdir(),"noise2.bin"))

xstar3 <- scan(file.path(tempdir(),"xstar3.dat"))
Unmasked03 <- unmask(maskedVectorToBeUnmasked=xstar3,
noisefile=file.path(tempdir(),"noise3.bin"))

xstar4 <- scan(file.path(tempdir(),"xstar4.dat"))
Unmasked04 <- unmask(maskedVectorToBeUnmasked=xstar4,
noisefile=file.path(tempdir(),"noise4.bin"))
# }
# NOT RUN {
# If the Data Provider has given the End-User the mean, standard deviation,
# or correlation matrix of the original variables
# }
# NOT RUN {
listOfMeansOfOriginalVariables<-list(mean(OriginalVar01), mean(OriginalVar02),
mean(OriginalVar03), mean(OriginalVar04))
listOfStandardDeviationsOfOriginalVariables<-list(sd(OriginalVar01), sd(OriginalVar02), 
sd(OriginalVar03), sd(OriginalVar04))
matrixOfCorrelationsOfOriginalVariables<-cor(cbind(OriginalVar01, OriginalVar02,
OriginalVar03, OriginalVar04))

test1 <- getSampleBasedOnUnmaskedData(maskedVectors =
list(xstar1, xstar2, xstar3, xstar4),
unmaskedVectors = list(Unmasked01$unmaskedVariable, 
Unmasked02$unmaskedVariable, Unmasked03$unmaskedVariable, 
Unmasked04$unmaskedVariable),
mu = listOfMeansOfOriginalVariables,
s = listOfStandardDeviationsOfOriginalVariables,  
rho_X = matrixOfCorrelationsOfOriginalVariables,
verbose = 2,
size = 1000)
# }
# NOT RUN {
# If the Data Provider has not, then these are estimated
# }
# NOT RUN {
test2 <- getSampleBasedOnUnmaskedData(meansOfNoises = 
list(Unmasked01$meanOfNoise, Unmasked02$meanOfNoise, 
Unmasked03$meanOfNoise, Unmasked04$meanOfNoise),
meansOfSquaredNoises = list(Unmasked01$meanOfSquaredNoise,
Unmasked02$meanOfSquaredNoise, Unmasked03$meanOfSquaredNoise,
Unmasked04$meanOfSquaredNoise),
maskedVectors = list(xstar1, xstar2, xstar3, xstar4),
unmaskedVectors = list(Unmasked01$unmaskedVariable, 
Unmasked02$unmaskedVariable, Unmasked03$unmaskedVariable, 
Unmasked04$unmaskedVariable),
verbose = 2,
size = 1000)

# }
# NOT RUN {
## alternatively - using batch versions
# }
# NOT RUN {
# Data Provider
set.seed(100)
data(FPI2002Data)
OriginalVar01 <- log(FPI2002Data$FPIflow2)
OriginalVar02 <- log(FPI2002Data$gdp_o)  
OriginalVar03 <- log(FPI2002Data$gdp_d) 
OriginalVar04 <- log(FPI2002Data$distw) 

testN <- nrow(FPI2002Data)

noise1<-rmulti(n=testN, mean=c(80, 100), sd=c(5,3), p=c(0.6, 0.4))
noise2<-rmulti(n=testN, mean=c(60, 90), sd=c(5,3), p=c(0.4, 0.6))
noise3<-rmulti(n=testN, mean=c(40, 110), sd=c(5,3), p=c(0.6, 0.4))
noise4<-rmulti(n=testN, mean=c(70, 100), sd=c(5,3), p=c(0.4, 0.6))

maskBatchOut <- maskBatch(listOfVectorsToBeMasked=
list(OriginalVar01,OriginalVar02,OriginalVar03,OriginalVar04),
listOfNoisefiles=list(file.path(tempdir(),"noise1.bin"),file.path(tempdir(),"noise2.bin"),
file.path(tempdir(),"noise3.bin"),file.path(tempdir(),"noise4.bin")),
listOfNoises=list(noise1,noise2,noise3,noise4),
listOfLowerBoundsAsGivenByProvider=
list(min(OriginalVar01),min(OriginalVar02),min(OriginalVar03),min(OriginalVar04)),
listofUpperBoundsAsGivenByProvider=
list(max(OriginalVar01),max(OriginalVar02),max(OriginalVar03),max(OriginalVar04)),
maxorder = 100, EPS = 1e-06)

dataFileNames <- list(file.path(tempdir(),"xstar1.dat"),file.path(tempdir(),"xstar2.dat"),
file.path(tempdir(),"xstar3.dat"),file.path(tempdir(),"xstar4.dat"))

for(i in 1:length(maskBatchOut)) {
write((maskBatchOut[[i]])$ystar, dataFileNames[[i]])
}
# }
# NOT RUN {
# End-User

dataFileNames <- list(file.path(tempdir(),"xstar1.dat"),file.path(tempdir(),"xstar2.dat"),
file.path(tempdir(),"xstar3.dat"),file.path(tempdir(),"xstar4.dat")) 
numberOfFiles <- length(dataFileNames)

maskedVectors <- list()
for(i in 1:numberOfFiles) {
maskedVectors[[i]] <- scan(dataFileNames[[i]]) 
}

# If the Data Provider has given the End-User the mean, standard deviation,
# or correlation matrix of the original variables
listOfMeansOfOriginalVariables<-list(mean(OriginalVar01), mean(OriginalVar02),
mean(OriginalVar03), mean(OriginalVar04))
listOfStandardDeviationsOfOriginalVariables<-list(sd(OriginalVar01), sd(OriginalVar02), 
sd(OriginalVar03), sd(OriginalVar04))
matrixOfCorrelationsOfOriginalVariables<-cor(cbind(OriginalVar01, OriginalVar02,
OriginalVar03, OriginalVar04))

unmaskAndGetSampleOut1 <- unmaskAndGetSampleBatch(listOfMaskedVectorsToBeUnmasked=maskedVectors,
                                    listOfNoisefiles=
                                    list(file.path(tempdir(),"noise1.bin"),
                                    file.path(tempdir(),"noise2.bin"),
                                    file.path(tempdir(),"noise3.bin"),
                                    file.path(tempdir(),"noise4.bin")),
                                    mu=listOfMeansOfOriginalVariables, 
                                    s=listOfStandardDeviationsOfOriginalVariables, 
                                    rho_X=matrixOfCorrelationsOfOriginalVariables,
                                    cores = 1, size=1000,
                                    verbose = 2,
                                    onlyUnmasked = FALSE)


# If the Data Provider has not, then these are estimated

unmaskAndGetSampleOut2 <- unmaskAndGetSampleBatch(listOfMaskedVectorsToBeUnmasked=maskedVectors,
                                    listOfNoisefiles=
                                    list(file.path(tempdir(),"noise1.bin"),
                                    file.path(tempdir(),"noise2.bin"),
                                    file.path(tempdir(),"noise3.bin"),
                                    file.path(tempdir(),"noise4.bin")),
                                    cores = 1, size=1000,
                                    verbose = 2,
                                    onlyUnmasked = FALSE)


# }