Nothing
R/CalcDissim.R
In DisimForMixed: Calculate Dissimilarity Matrix for Dataset with Mixed Attributes
#' Descretize Quantitative Variables.
#'
#' Takes in a data frame which contains only Quantitative varables in columns. Standadize the variables.
#' Discretize quantitative variables and returns discretized quantitative variables. Discretization was performed by
#' equal width bining algorithm.
#' @param myDataQuant A data frame which includes quantitative variables in columns.
#' @param noice Noice indicator. If noice = TRUE data standerdization is done by
#' deviding the difference between data point and median of the variable by the range of the variable.
#' If noice = FALSE data standerdization is done by deviding the difference between data point and
#' mean of the variable by the standard deviation of the variable.
#' @return A data frame consists of discretized quantitative variables.
#' @examples
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' @import stats
#' @export
discretizeQuant <- function(myDataQuant, noice = TRUE){
i = 1
noOfObj <-nrow(myDataQuant)
noOfVar <- ncol(myDataQuant)
NormData <- data.frame(data = NA)
if(noice == TRUE){
for(i in 1:ncol(myDataQuant)){
NormData_i <- (myDataQuant[,i] - stats::median(myDataQuant[,i]))/(max(myDataQuant[,i]) - min(myDataQuant[,i]))
NormData <- cbind(NormData, NormData_i)
}
}
if(noice == FALSE){
for(i in 1:ncol(myDataQuant)){
NormData_i <- (myDataQuant[,i] - mean(myDataQuant[,i]))/ stats::sd(myDataQuant[,i])
NormData <- cbind(NormData, NormData_i)
}
}
NormData <- data.frame(NormData[,-1])
colnames(NormData) <- colnames(myDataQuant)
#decretize by equal no of Intervals
dectrTzed <- data.frame()
dectrTzedFac <- matrix(data = NA, ncol = noOfVar, nrow = noOfObj)
noOfInt <- ceiling(sqrt(noOfObj))
IntWidth <- (max(NormData[,i]) - min(NormData[,i])) / noOfInt
for(i in 1:ncol(NormData)){
for (j in 1:noOfObj){
k = 1
while(k <= noOfInt){
if(NormData[j,i] <= min(NormData[,i]) + k*IntWidth){
dectrTzed[j,i] <- k
break
}
if(k == noOfInt)
dectrTzed[j,i] <- k
k = k + 1
}
}
}
i = 1
#Find Significance
for(i in 1:ncol(myDataQuant)){
dectrTzedFac[,i] <- as.factor(dectrTzed[,i])
}
dectrTzedFac <- data.frame(dectrTzedFac)
colnames(dectrTzedFac) <- colnames(myDataQuant)
return(dectrTzedFac)
}
#' Calculate Conditional Probabilities.
#'
#' Takes in a data frame which contains only qualitative variables. Discretized quantitative variables
#' , a mixture of qualitative variables and discretized quantitative variables are also accepted.
#' Calculates conditional probabilities for each pair of attribute values in the data frame.
#' Returns a data frame consists of J, A, B and C in columns where Pr(A|B) = C and J is the column number in the input
#' data frame corresponding to the values in A.
#' @param myDataAll A data frame which includes qualitative variables OR discretized quantitative variables
#' OR a mixture of qualitative variables and discretized quantitative variables in columns.
#' @return A data frame with four columns J, A, B and C in columns where Pr(A|B) = C and J is the column number in the input
#' data frame corresponding to the values in A.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' CalcForQuali <- calcCondProb(QualiVars)
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' CalcForQuant <- calcCondProb(Discretized)
#' AllQualQuant <- data.frame(QualiVars, Discretized)
#' CalcForAll <- calcCondProb(AllQualQuant)
#' @export
calcCondProb <- function(myDataAll){
ProbData <- data.frame()
NoOfVars <- ncol(myDataAll)
for(i in 1:NoOfVars){
levelsOfDataFirst <- data.frame(levels(as.factor(myDataAll[,i])))
NofFirst <- nrow(levelsOfDataFirst)
j = 1
while(j <= NoOfVars){
if(j == i)
j = j + 1
if(j > NoOfVars)
break
levelsOfDataSec <- data.frame(levels(as.factor(myDataAll[,j])))
NofSec <- nrow(levelsOfDataSec)
for(k in 1:NofFirst){
filteredfRST = myDataAll[myDataAll[,i] == levelsOfDataFirst[k,1], ]
denomVal <- nrow(filteredfRST)
for(a in 1:NofSec){
filteredSec <- filteredfRST[filteredfRST[,j] == levelsOfDataSec[a,1], ]
if (is.na(filteredSec[1,1]) == FALSE){
numeVal <- nrow(filteredSec)
}else
numeVal <- 0
jVal = j
occur <- levelsOfDataSec[a,1]
given <- levelsOfDataFirst[k,1]
condProbVal <- numeVal/denomVal
newRowToBind <- data.frame(jVal, occur, given, condProbVal)
ProbData <- rbind(ProbData,newRowToBind)
}
}
j = j + 1
}
}
return(ProbData)
}
#' Calculate Distance Between given Attribute Values by considering only a pair of attributes.
#'
#' Takes in two lists Ai and Aj, representing values of two attributes, two values x and y from Ai.
#' Quantitative attributes are accepted only after descretization. Calculates distance between x and y
#' for Aj with respect to Ai.
#' @param Ai A list consisting values of a selected attribute
#' @param Aj A list consisting values of another selected attribute
#' @param x Value from Ai
#' @param y Another value from Ai
#' @details findMax is the implementation of find_max() function proposed by Ahmad & Dey (2007).
#' See Ahmad & Dey (2007) for more datails.
#' @return distance between x and y for Aj with respect to Ai.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' Attrib_i <- c("A","B","A","C")
#' Attrib_j <- c("Q","Q","R","Q")
#' xVal <- "A"
#' yVal <- "B"
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' library(dplyr)
#' distBetXY <- findMax(Attrib_i,Attrib_j,xVal,yVal)
#' @export
findMax <- function(Ai, Aj, x, y){
dist_xy = 0
levelsOfAj <- data.frame(levels(as.factor(Aj)))
NofAj <- nrow(levelsOfAj)
ConditioanalProbs <- calcCondProb(data.frame(Ai,Aj))
filteredCP = data.frame()
for (i in 1:NofAj){
filteredCPi <- dplyr::filter(ConditioanalProbs, ConditioanalProbs$occur == as.character(levelsOfAj[i,1]) & (ConditioanalProbs$given == x | ConditioanalProbs$given == y) & ConditioanalProbs$jVal == 2)
filteredCP <- rbind(filteredCP,filteredCPi)
}
for(t in 1:NofAj){
jGivx = 100
jGivy = 100
for (k in 1:nrow(filteredCP)){
if(as.character(filteredCP$occur[k]) == as.character(levelsOfAj[t,1]) & as.character(filteredCP$given[k]) == x){
jGivx <- filteredCP$condProbVal[k]
}
if(as.character(filteredCP$occur[k]) == as.character(levelsOfAj[t,1]) & as.character(filteredCP$given[k]) == y){
jGivy <- filteredCP$condProbVal[k]
}
if(jGivx != 100 & jGivy != 100){
break
}
}
if(jGivx >= jGivy){
dist_xy = dist_xy + jGivx
}else{
dist_xy = dist_xy + jGivy
}
}
dist_xy <- dist_xy - 1
return(dist_xy)
}
#' Calculate Distance Between Attribute Values.
#'
#' Takes in a data frame which contains only qualitative variables. Discretized quantitative variables
#' , a mixture of qualitative variables and discretized quantitative variables are also accepted.
#' Calculates distance between each pair of attribute values for a given attribute. This calculation
#' is done according to the method proposed by Ahmad & Dey (2007).
#' @param myDataAll A data frame which includes qualitative variables OR discretized quantitative variables
#' OR a mixture of qualitative variables and discretized quantitative variables in columns.
#' @details distBetPairs is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to find the distance between two catogorical values corresponding to a qualitative
#' variable. This distance measure considers distribution of values in the data set.
#' This function is also used to find the distance between discretized values
#' corresponding to quantitative variables which are used in calculating the significance
#' of quantitative attributes. See Ahmad & Dey (2007) for more datails.
#' @return A data frame with four columns J, A, B and C in columns where Distance(A, B) = C
#' and J is the column number in the input data frame corresponding to the values in A.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' library(dplyr)
#' distForQuali <- distBetPairs(QualiVars)
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' distForQuant <- distBetPairs(Discretized)
#' AllQualQuant <- data.frame(QualiVars, Discretized)
#' distForAll <- distBetPairs(AllQualQuant)
#' @export
distBetPairs <- function(myDataAll){
DistData <- data.frame()
NoOfVars <- ncol(myDataAll)
for(i in 1:NoOfVars){
levelsOfDataFirst <- data.frame(levels(as.factor(myDataAll[,i])))
NofFirst <- nrow(levelsOfDataFirst)
jStop <- NofFirst-1
for (j in 1:jStop){
aStart <- j + 1
for (a in aStart:NofFirst){
SumProb <- 0
k=1
while(k <= NoOfVars){
if(k != i){
distXY <- findMax(as.character(myDataAll[,i]), as.character(myDataAll[,k]), as.character(levelsOfDataFirst[j,1]), as.character(levelsOfDataFirst[a,1]))
SumProb <- SumProb + distXY
}
k = k + 1
}
VarIndex <- i
xVal <- levelsOfDataFirst[j,1]
yVal <- levelsOfDataFirst[a,1]
DistBetPair <- SumProb/(NoOfVars-1)
newRowToBind <- data.frame(VarIndex, xVal, yVal)
newRowToBind <- data.frame(VarIndex, xVal, yVal, DistBetPair)
DistData <- rbind(DistData, newRowToBind)
}
}
}
return(DistData)
}
#' Calculate Significance of Quantitative Attributes.
#'
#' Takes in two data frames where first contains only qualitative attributes and the other
#' contains only quantitative attributes. Function calculates significance of quantitative
#' attributes based on the method proposed by Ahmad & Dey (2007).
#' @param myDataQuali A data frame which includes only qualitative variables in columns.
#' @param myDataQuant A data frame which includes only quantitative variables in columns.
#' @details signifOfQuantVars is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to calculate the significance of quantitative attributes. Signinficance of an attribute is an
#' important fact to consider in the process of clustering. To calculate the significance
#' quantitative attributes are discreized first. These significace values are used in calculating
#' distance between any two numeric values of aquantitative attribute. See Ahmad & Dey (2007) for
#' more datails.
#' @return A data frame with two columns A and B where A represents variable number and B
#' represents significane of corresponding variable.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' SigOfQuant <- signifOfQuantVars(QualiVars, QuantVars)
#' @export
signifOfQuantVars <- function(myDataQuali, myDataQuant){
DiscretQuant <- discretizeQuant(myDataQuant)
allData <- data.frame(myDataQuali, DiscretQuant)
DistForAll <- distBetPairs(allData)
QuantVarStart <- ncol(myDataQuali) + 1
LoopStopVal <- ncol(allData)
SignfData = data.frame()
ct = 1
for(i in QuantVarStart:LoopStopVal){
VarIndex = DistForAll[i,1]
SigDataForQuant <- dplyr::filter(DistForAll, VarIndex == i)
SignfData[ct,1] <- ct
SignfData[ct,2] <- mean(SigDataForQuant$DistBetPair)
ct = ct + 1
}
return(SignfData)
}
#' Calculate Dissimilarity Matrix for Mixed Attributes.
#'
#' Takes in two data frames where first contains only qualitative attributes and the other
#' contains only quantitative attributes. Function calculates the dissimilarity matrix
#' based on the method proposed by Ahmad & Dey (2007).
#' @param myDataQuali A data frame which includes only qualitative variables in columns.
#' @param myDataQuant A data frame which includes only quantitative variables in columns.
#' @details calcDissimMat is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to calculate the dissimilarity matrix at the presence of both qualitative and quantitative
#' attributes. This approach finds dissimilarity of qualitative and quantitative attributes seperately
#' and the final dissimilarity matrix is formed by combining both. See Ahmad & Dey (2007) for
#' more datails.
#' @return A dissimilarity matrix. This can be used as an input to pam, fanny, agnes and diana functions.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C","C","A"), Qlvar2 = c("Q","Q","R","Q","R","Q"))
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5,2.8,3.1), Qnvar2 = c(4.8,2,1.1,5.8,3.1,2.2))
#' DisSimMatCalcd <- calcDissimMat(QualiVars, QuantVars)
#'
#' agnesClustering <- cluster::agnes(DisSimMatCalcd, diss = TRUE, method = "ward")
#' silWidths <- cluster::silhouette(cutree(agnesClustering, k = 2), DisSimMatCalcd)
#' mean(silWidths[,3])
#' plot(agnesClustering)
#'
#' PAMClustering <- cluster::pam(DisSimMatCalcd, k=2, diss = TRUE)
#' silWidths <- cluster::silhouette(PAMClustering, DisSimMatCalcd)
#' plot(silWidths)
#' @export
calcDissimMat <- function(myDataQuali, myDataQuant){
#library(dplyr)
dissimForQuant <- data.frame()
dissimForQuali <- data.frame()
dissimForAll <- data.frame()
disSimForAttr <- numeric(0)
NoOfObj <- nrow(myDataQuali)
NoClolQuant <- ncol(myDataQuant)
NoClolQuali <- ncol(myDataQuali)
SignfDataFound <- signifOfQuantVars(myDataQuali, myDataQuant)
for (a in 1:NoOfObj){
for (b in 1:NoOfObj){
for (i in 1:NoClolQuant){
disSimForAttr[i] <- (SignfDataFound[i,2]*(myDataQuant[a,i] - myDataQuant[b,i]))^2
}
dissimForQuant[a,b] <- sum(disSimForAttr)
}
}
disSimForAttr <- numeric(0)
DistanceData <- distBetPairs(myDataQuali)
for (a in 1:NoOfObj){
for (b in 1:NoOfObj){
for (i in 1:NoClolQuali){
DistanceFound <- dplyr::filter(DistanceData, (DistanceData$xVal == as.character(myDataQuali[a,i]) & DistanceData$yVal == as.character(myDataQuali[b,i])) | (DistanceData$xVal == as.character(myDataQuali[b,i]) & DistanceData$yVal == as.character(myDataQuali[a,i])))
if(nrow(DistanceFound) == 0){
DistValUniq = 0
}else
DistValUniq <- DistanceFound$DistBetPair
disSimForAttr[i] <- DistValUniq ^ 2
}
dissimForQuali[a,b] <- sum(disSimForAttr)
}
}
dissimForAll <- dissimForQuant[,] + dissimForQuali[,]
agnesClustering <- cluster::agnes(dissimForAll, diss = TRUE, method = "ward")
return(dissimForAll)
}
Try the DisimForMixed package in your browser
Any scripts or data that you put into this service are public.
DisimForMixed documentation built on May 1, 2019, 9:25 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
#' Descretize Quantitative Variables.
#'
#' Takes in a data frame which contains only Quantitative varables in columns. Standadize the variables.
#' Discretize quantitative variables and returns discretized quantitative variables. Discretization was performed by
#' equal width bining algorithm.
#' @param myDataQuant A data frame which includes quantitative variables in columns.
#' @param noice Noice indicator. If noice = TRUE data standerdization is done by
#' deviding the difference between data point and median of the variable by the range of the variable.
#' If noice = FALSE data standerdization is done by deviding the difference between data point and
#' mean of the variable by the standard deviation of the variable.
#' @return A data frame consists of discretized quantitative variables.
#' @examples
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' @import stats
#' @export
discretizeQuant <- function(myDataQuant, noice = TRUE){
i = 1
noOfObj <-nrow(myDataQuant)
noOfVar <- ncol(myDataQuant)
NormData <- data.frame(data = NA)
if(noice == TRUE){
for(i in 1:ncol(myDataQuant)){
NormData_i <- (myDataQuant[,i] - stats::median(myDataQuant[,i]))/(max(myDataQuant[,i]) - min(myDataQuant[,i]))
NormData <- cbind(NormData, NormData_i)
}
}
if(noice == FALSE){
for(i in 1:ncol(myDataQuant)){
NormData_i <- (myDataQuant[,i] - mean(myDataQuant[,i]))/ stats::sd(myDataQuant[,i])
NormData <- cbind(NormData, NormData_i)
}
}
NormData <- data.frame(NormData[,-1])
colnames(NormData) <- colnames(myDataQuant)
#decretize by equal no of Intervals
dectrTzed <- data.frame()
dectrTzedFac <- matrix(data = NA, ncol = noOfVar, nrow = noOfObj)
noOfInt <- ceiling(sqrt(noOfObj))
IntWidth <- (max(NormData[,i]) - min(NormData[,i])) / noOfInt
for(i in 1:ncol(NormData)){
for (j in 1:noOfObj){
k = 1
while(k <= noOfInt){
if(NormData[j,i] <= min(NormData[,i]) + k*IntWidth){
dectrTzed[j,i] <- k
break
}
if(k == noOfInt)
dectrTzed[j,i] <- k
k = k + 1
}
}
}
i = 1
#Find Significance
for(i in 1:ncol(myDataQuant)){
dectrTzedFac[,i] <- as.factor(dectrTzed[,i])
}
dectrTzedFac <- data.frame(dectrTzedFac)
colnames(dectrTzedFac) <- colnames(myDataQuant)
return(dectrTzedFac)
}
#' Calculate Conditional Probabilities.
#'
#' Takes in a data frame which contains only qualitative variables. Discretized quantitative variables
#' , a mixture of qualitative variables and discretized quantitative variables are also accepted.
#' Calculates conditional probabilities for each pair of attribute values in the data frame.
#' Returns a data frame consists of J, A, B and C in columns where Pr(A|B) = C and J is the column number in the input
#' data frame corresponding to the values in A.
#' @param myDataAll A data frame which includes qualitative variables OR discretized quantitative variables
#' OR a mixture of qualitative variables and discretized quantitative variables in columns.
#' @return A data frame with four columns J, A, B and C in columns where Pr(A|B) = C and J is the column number in the input
#' data frame corresponding to the values in A.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' CalcForQuali <- calcCondProb(QualiVars)
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' CalcForQuant <- calcCondProb(Discretized)
#' AllQualQuant <- data.frame(QualiVars, Discretized)
#' CalcForAll <- calcCondProb(AllQualQuant)
#' @export
calcCondProb <- function(myDataAll){
ProbData <- data.frame()
NoOfVars <- ncol(myDataAll)
for(i in 1:NoOfVars){
levelsOfDataFirst <- data.frame(levels(as.factor(myDataAll[,i])))
NofFirst <- nrow(levelsOfDataFirst)
j = 1
while(j <= NoOfVars){
if(j == i)
j = j + 1
if(j > NoOfVars)
break
levelsOfDataSec <- data.frame(levels(as.factor(myDataAll[,j])))
NofSec <- nrow(levelsOfDataSec)
for(k in 1:NofFirst){
filteredfRST = myDataAll[myDataAll[,i] == levelsOfDataFirst[k,1], ]
denomVal <- nrow(filteredfRST)
for(a in 1:NofSec){
filteredSec <- filteredfRST[filteredfRST[,j] == levelsOfDataSec[a,1], ]
if (is.na(filteredSec[1,1]) == FALSE){
numeVal <- nrow(filteredSec)
}else
numeVal <- 0
jVal = j
occur <- levelsOfDataSec[a,1]
given <- levelsOfDataFirst[k,1]
condProbVal <- numeVal/denomVal
newRowToBind <- data.frame(jVal, occur, given, condProbVal)
ProbData <- rbind(ProbData,newRowToBind)
}
}
j = j + 1
}
}
return(ProbData)
}
#' Calculate Distance Between given Attribute Values by considering only a pair of attributes.
#'
#' Takes in two lists Ai and Aj, representing values of two attributes, two values x and y from Ai.
#' Quantitative attributes are accepted only after descretization. Calculates distance between x and y
#' for Aj with respect to Ai.
#' @param Ai A list consisting values of a selected attribute
#' @param Aj A list consisting values of another selected attribute
#' @param x Value from Ai
#' @param y Another value from Ai
#' @details findMax is the implementation of find_max() function proposed by Ahmad & Dey (2007).
#' See Ahmad & Dey (2007) for more datails.
#' @return distance between x and y for Aj with respect to Ai.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' Attrib_i <- c("A","B","A","C")
#' Attrib_j <- c("Q","Q","R","Q")
#' xVal <- "A"
#' yVal <- "B"
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' library(dplyr)
#' distBetXY <- findMax(Attrib_i,Attrib_j,xVal,yVal)
#' @export
findMax <- function(Ai, Aj, x, y){
dist_xy = 0
levelsOfAj <- data.frame(levels(as.factor(Aj)))
NofAj <- nrow(levelsOfAj)
ConditioanalProbs <- calcCondProb(data.frame(Ai,Aj))
filteredCP = data.frame()
for (i in 1:NofAj){
filteredCPi <- dplyr::filter(ConditioanalProbs, ConditioanalProbs$occur == as.character(levelsOfAj[i,1]) & (ConditioanalProbs$given == x | ConditioanalProbs$given == y) & ConditioanalProbs$jVal == 2)
filteredCP <- rbind(filteredCP,filteredCPi)
}
for(t in 1:NofAj){
jGivx = 100
jGivy = 100
for (k in 1:nrow(filteredCP)){
if(as.character(filteredCP$occur[k]) == as.character(levelsOfAj[t,1]) & as.character(filteredCP$given[k]) == x){
jGivx <- filteredCP$condProbVal[k]
}
if(as.character(filteredCP$occur[k]) == as.character(levelsOfAj[t,1]) & as.character(filteredCP$given[k]) == y){
jGivy <- filteredCP$condProbVal[k]
}
if(jGivx != 100 & jGivy != 100){
break
}
}
if(jGivx >= jGivy){
dist_xy = dist_xy + jGivx
}else{
dist_xy = dist_xy + jGivy
}
}
dist_xy <- dist_xy - 1
return(dist_xy)
}
#' Calculate Distance Between Attribute Values.
#'
#' Takes in a data frame which contains only qualitative variables. Discretized quantitative variables
#' , a mixture of qualitative variables and discretized quantitative variables are also accepted.
#' Calculates distance between each pair of attribute values for a given attribute. This calculation
#' is done according to the method proposed by Ahmad & Dey (2007).
#' @param myDataAll A data frame which includes qualitative variables OR discretized quantitative variables
#' OR a mixture of qualitative variables and discretized quantitative variables in columns.
#' @details distBetPairs is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to find the distance between two catogorical values corresponding to a qualitative
#' variable. This distance measure considers distribution of values in the data set.
#' This function is also used to find the distance between discretized values
#' corresponding to quantitative variables which are used in calculating the significance
#' of quantitative attributes. See Ahmad & Dey (2007) for more datails.
#' @return A data frame with four columns J, A, B and C in columns where Distance(A, B) = C
#' and J is the column number in the input data frame corresponding to the values in A.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' library(dplyr)
#' distForQuali <- distBetPairs(QualiVars)
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' Discretized <- discretizeQuant(QuantVars)
#' distForQuant <- distBetPairs(Discretized)
#' AllQualQuant <- data.frame(QualiVars, Discretized)
#' distForAll <- distBetPairs(AllQualQuant)
#' @export
distBetPairs <- function(myDataAll){
DistData <- data.frame()
NoOfVars <- ncol(myDataAll)
for(i in 1:NoOfVars){
levelsOfDataFirst <- data.frame(levels(as.factor(myDataAll[,i])))
NofFirst <- nrow(levelsOfDataFirst)
jStop <- NofFirst-1
for (j in 1:jStop){
aStart <- j + 1
for (a in aStart:NofFirst){
SumProb <- 0
k=1
while(k <= NoOfVars){
if(k != i){
distXY <- findMax(as.character(myDataAll[,i]), as.character(myDataAll[,k]), as.character(levelsOfDataFirst[j,1]), as.character(levelsOfDataFirst[a,1]))
SumProb <- SumProb + distXY
}
k = k + 1
}
VarIndex <- i
xVal <- levelsOfDataFirst[j,1]
yVal <- levelsOfDataFirst[a,1]
DistBetPair <- SumProb/(NoOfVars-1)
newRowToBind <- data.frame(VarIndex, xVal, yVal)
newRowToBind <- data.frame(VarIndex, xVal, yVal, DistBetPair)
DistData <- rbind(DistData, newRowToBind)
}
}
}
return(DistData)
}
#' Calculate Significance of Quantitative Attributes.
#'
#' Takes in two data frames where first contains only qualitative attributes and the other
#' contains only quantitative attributes. Function calculates significance of quantitative
#' attributes based on the method proposed by Ahmad & Dey (2007).
#' @param myDataQuali A data frame which includes only qualitative variables in columns.
#' @param myDataQuant A data frame which includes only quantitative variables in columns.
#' @details signifOfQuantVars is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to calculate the significance of quantitative attributes. Signinficance of an attribute is an
#' important fact to consider in the process of clustering. To calculate the significance
#' quantitative attributes are discreized first. These significace values are used in calculating
#' distance between any two numeric values of aquantitative attribute. See Ahmad & Dey (2007) for
#' more datails.
#' @return A data frame with two columns A and B where A represents variable number and B
#' represents significane of corresponding variable.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C"), Qlvar2 = c("Q","Q","R","Q"))
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5), Qnvar2 = c(4.8,2,1.1,5.8))
#' SigOfQuant <- signifOfQuantVars(QualiVars, QuantVars)
#' @export
signifOfQuantVars <- function(myDataQuali, myDataQuant){
DiscretQuant <- discretizeQuant(myDataQuant)
allData <- data.frame(myDataQuali, DiscretQuant)
DistForAll <- distBetPairs(allData)
QuantVarStart <- ncol(myDataQuali) + 1
LoopStopVal <- ncol(allData)
SignfData = data.frame()
ct = 1
for(i in QuantVarStart:LoopStopVal){
VarIndex = DistForAll[i,1]
SigDataForQuant <- dplyr::filter(DistForAll, VarIndex == i)
SignfData[ct,1] <- ct
SignfData[ct,2] <- mean(SigDataForQuant$DistBetPair)
ct = ct + 1
}
return(SignfData)
}
#' Calculate Dissimilarity Matrix for Mixed Attributes.
#'
#' Takes in two data frames where first contains only qualitative attributes and the other
#' contains only quantitative attributes. Function calculates the dissimilarity matrix
#' based on the method proposed by Ahmad & Dey (2007).
#' @param myDataQuali A data frame which includes only qualitative variables in columns.
#' @param myDataQuant A data frame which includes only quantitative variables in columns.
#' @details calcDissimMat is an implementtion of the method proposed by Ahmad & Dey (2007)
#' to calculate the dissimilarity matrix at the presence of both qualitative and quantitative
#' attributes. This approach finds dissimilarity of qualitative and quantitative attributes seperately
#' and the final dissimilarity matrix is formed by combining both. See Ahmad & Dey (2007) for
#' more datails.
#' @return A dissimilarity matrix. This can be used as an input to pam, fanny, agnes and diana functions.
#' @references Ahmad, A., & Dey, L. (2007). A k-mean clustering algorithm for mixed numeric and categorical data. Data & Knowledge Engineering, 63(2), 503-527.
#' @examples
#' QualiVars <- data.frame(Qlvar1 = c("A","B","A","C","C","A"), Qlvar2 = c("Q","Q","R","Q","R","Q"))
#' QuantVars <- data.frame(Qnvar1 = c(1.5,3.2,4.9,5,2.8,3.1), Qnvar2 = c(4.8,2,1.1,5.8,3.1,2.2))
#' DisSimMatCalcd <- calcDissimMat(QualiVars, QuantVars)
#'
#' agnesClustering <- cluster::agnes(DisSimMatCalcd, diss = TRUE, method = "ward")
#' silWidths <- cluster::silhouette(cutree(agnesClustering, k = 2), DisSimMatCalcd)
#' mean(silWidths[,3])
#' plot(agnesClustering)
#'
#' PAMClustering <- cluster::pam(DisSimMatCalcd, k=2, diss = TRUE)
#' silWidths <- cluster::silhouette(PAMClustering, DisSimMatCalcd)
#' plot(silWidths)
#' @export
calcDissimMat <- function(myDataQuali, myDataQuant){
#library(dplyr)
dissimForQuant <- data.frame()
dissimForQuali <- data.frame()
dissimForAll <- data.frame()
disSimForAttr <- numeric(0)
NoOfObj <- nrow(myDataQuali)
NoClolQuant <- ncol(myDataQuant)
NoClolQuali <- ncol(myDataQuali)
SignfDataFound <- signifOfQuantVars(myDataQuali, myDataQuant)
for (a in 1:NoOfObj){
for (b in 1:NoOfObj){
for (i in 1:NoClolQuant){
disSimForAttr[i] <- (SignfDataFound[i,2]*(myDataQuant[a,i] - myDataQuant[b,i]))^2
}
dissimForQuant[a,b] <- sum(disSimForAttr)
}
}
disSimForAttr <- numeric(0)
DistanceData <- distBetPairs(myDataQuali)
for (a in 1:NoOfObj){
for (b in 1:NoOfObj){
for (i in 1:NoClolQuali){
DistanceFound <- dplyr::filter(DistanceData, (DistanceData$xVal == as.character(myDataQuali[a,i]) & DistanceData$yVal == as.character(myDataQuali[b,i])) | (DistanceData$xVal == as.character(myDataQuali[b,i]) & DistanceData$yVal == as.character(myDataQuali[a,i])))
if(nrow(DistanceFound) == 0){
DistValUniq = 0
}else
DistValUniq <- DistanceFound$DistBetPair
disSimForAttr[i] <- DistValUniq ^ 2
}
dissimForQuali[a,b] <- sum(disSimForAttr)
}
}
dissimForAll <- dissimForQuant[,] + dissimForQuali[,]
agnesClustering <- cluster::agnes(dissimForAll, diss = TRUE, method = "ward")
return(dissimForAll)
}
Try the DisimForMixed package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.