Nothing
R/input_validation.R
In clustord: Cluster Ordinal Data via Proportional Odds or Ordered Stereotype
Defines functions
extract.covs
convert.model.row.to.column
check.formula
check.factors
validate.inputs
#' @importFrom methods is
validate.inputs <- function(formula, model,
nclus.row=NULL,nclus.column=NULL,
long.df,
initvect=NULL,
pi.init=NULL, kappa.init=NULL,
EM.control=default.EM.control(),
optim.method="L-BFGS-B",
constraint_sum_zero=TRUE,
start_from_simple_model=TRUE,
parallel_starts=FALSE,
nstarts=5,
verbose=TRUE) {
## Note the double-& and double-| which stops the later parts being checked
## if the earlier parts are false
if (!is(formula, "formula")) stop("formula must be a valid formula.")
## Check that model is valid
if (!is.character(model) || !is.vector(model) || length(model) != 1) stop("model must be a string, 'OSM' or 'POM' or 'Binary'.")
if (!(model %in% c("OSM","POM","Binary"))) stop("model must be either 'OSM' or POM' for the ordered stereotype and proportional odds models, or 'Binary' for the binary model.")
## Check that clustering settings are valid (later, in check.formula, will
## make sure user has provide nclus.row or nclus.column or both, depending
## on formula supplied)
if (!is.null(nclus.row)) {
if (!is.vector(nclus.row) || length(nclus.row) != 1 || nclus.row <= 1 ||
nclus.row %% 1 != 0 || is.na(nclus.row)) {
stop("nclus.row must be an integer, from 2 to the number of rows/observations in the data.")
}
}
if (!is.null(nclus.column)) {
if (!is.vector(nclus.column) || length(nclus.column) != 1 ||
nclus.column <= 1 || nclus.column %% 1 != 0 || is.na(nclus.column)) {
stop("nclus.column must be an integer, from 2 to the number of columns/questions in the data.")
}
}
if (is.null(long.df)) stop("long.df cannot be null.")
if (!is.data.frame(long.df)) stop("long.df must be a data frame.")
if (length(long.df) < 3) stop("long.df must have at least 3 columns, Y and ROW and COL.")
names.df <- names(long.df)
# Convert the Y, ROW and COL variables to upper case if they are in the df but
# in the wrong case
if ("y" %in% names.df) names.df[which(names.df == "y")] <- "Y"
if ("ROW" %in% toupper(names.df)) names.df[which(toupper(names.df) == "ROW")] <- "ROW"
if ("COL" %in% toupper(names.df)) names.df[which(toupper(names.df) == "COL")] <- "COL"
if (!("Y" %in% names.df)) stop("long.df must have a column named 'Y' which contains the response values.")
if (!("ROW" %in% names.df)) stop("long.df must have a column named 'ROW' which indicates what observation (row in the data matrix) each value of Y corresponds to.")
if (!("COL" %in% names.df)) stop("long.df must have a column named 'COL' which indicates what variable (column in the data matrix) each value of Y corresponds to.")
if (length(which(names.df == "Y")) > 1) stop("long.df should only have one column named 'Y'.")
if (length(which(names.df == "ROW")) > 1) stop("long.df should only have one column named 'ROW'.")
if (length(which(names.df == "COL")) > 1) stop("long.df should only have one column named 'COL'.")
if (!is.factor(long.df$Y)) stop("long.df$Y must be a factor.")
if (any(is.na(long.df$Y))) stop("long.df$Y has missing values (NA). Please delete these rows and try again.")
if (is.list(long.df$Y) || any(sapply(long.df$Y,is.list)) ||
any(sapply(long.df$Y,is.infinite))) stop("long.df$Y is a list, or has list elements or infinite elements. long.df$Y should be a factor with q levels.")
if (model == "Binary" && length(unique(long.df$Y)) > 2) stop("For the Binary model, long.df$Y should only have 2 possible values.")
if (!is.factor(long.df$ROW) &&
(is.list(long.df$ROW) || any(sapply(long.df$ROW,is.list)) || any(is.na(long.df$ROW)) ||
any(sapply(long.df$ROW,is.infinite)) || any(long.df$ROW %% 1 != 0) ||
any(long.df$ROW < 1) || all(long.df$ROW > 1))) stop("long.df$ROW must be a factor or integers from 1 to the number of observations, i.e. the number of rows in the original data matrix.")
if (!is.factor(long.df$COL) &&
(is.list(long.df$COL) || any(sapply(long.df$COL,is.list)) || any(is.na(long.df$COL)) ||
any(sapply(long.df$COL,is.infinite)) || any(long.df$COL %% 1 != 0) ||
any(long.df$COL < 1) || all(long.df$COL > 1))) stop("long.df$COL must be a factor or integers from 1 to the number of variables, i.e. the number of columns in the original data matrix.")
if (any(table(long.df[,c("ROW","COL")]) > 1)) stop("Each element from the original data matrix must correspond to no more than 1 row in long.df.")
cov_idxs <- which(!(names(long.df) %in% c("Y","ROW","COL")))
for (j in cov_idxs) {
col <- long.df[,j]
non_na_col <- col[!is.na(col)]
if (length(unique(non_na_col)) == 1) stop(paste("Covariate",names(long.df)[j],"only takes one non-missing value for all entries of the data matrix. Please remove this covariate before continuing."))
}
if (!is.null(nclus.row) && nclus.row >= max(as.numeric(long.df$ROW))) stop("nclus.row must be smaller than the maximum value of long.df$ROW.")
if (!is.null(nclus.column) && nclus.column >= max(as.numeric(long.df$COL))) stop("nclus.column must be smaller than the maximum value of long.df$COL.")
if (!is.null(initvect)) {
if (!is.vector(initvect) || !is.numeric(initvect) || any(is.na(initvect)) ||
any(is.infinite(initvect))) stop("If supplied, initvect must be a numeric vector with finite values.")
}
if (!is.null(pi.init)) {
if (!is.vector(pi.init) || !is.numeric(pi.init) || any(is.na(pi.init)) ||
any(pi.init < 0) || any(pi.init > 1)) stop("If supplied, pi.init must be a vector of numbers between 0 and 1.")
if (length(pi.init) != nclus.row || abs(sum(pi.init) - 1) > 1e-12) stop("pi.init must be the same length as the number of row clusters, and must add up to 1")
}
if (!is.null(kappa.init)) {
if (!is.vector(kappa.init) || !is.numeric(kappa.init) || any(is.na(kappa.init)) ||
any(kappa.init < 0) | any(kappa.init > 1)) stop("If supplied, kappa.init must be a vector of numbers between 0 and 1.")
if (length(kappa.init) != nclus.column || abs(sum(kappa.init) - 1) > 1e-12) stop("kappa.init must be the same length as the number of column clusters, and must add up to 1")
}
if (!is.logical(constraint_sum_zero) || !is.vector(constraint_sum_zero) ||
length(constraint_sum_zero) != 1 || is.na(constraint_sum_zero)) stop("constraint_sum_zero must be TRUE or FALSE.")
if (!is.logical(start_from_simple_model) || !is.vector(start_from_simple_model) ||
length(start_from_simple_model) != 1 || is.na(start_from_simple_model)) stop("start_from_simple_model must be TRUE or FALSE.")
if (!is.null(nstarts)) {
if (!is.vector(nstarts) || !is.numeric(nstarts) || length(nstarts) != 1 ||
nstarts < 0 || nstarts %% 1 != 0) stop("If supplied, nstarts must be a positive integer.")
}
if (!(parallel_starts %in% c(TRUE,FALSE))) {
stop("parallel_starts must be TRUE or FALSE.")
}
if (!is.null(EM.control) & (!is.list(EM.control) || length(EM.control) == 0 || length(EM.control) > 9 ||
!all(names(EM.control) %in% c("EMcycles","EMlikelihoodtol","EMparamtol",
"paramstopping","startEMcycles","keepallparams",
"epsilon", "rerunestepbeforelli",
"uselatestlli")))) {
stop("If supplied, EM.control must be a list of control parameters for the EM algorithm. Please see the manual for more info.")
}
if (is.null(optim.method) || !is.character(optim.method) || !is.vector(optim.method) ||
length(optim.method) != 1 || !(optim.method %in% c("Nelder-Mead","BFGS","CG","L-BFGS-B"))) stop("If supplied, optim.method must be one of the valid methods for optim, 'Nelder-Mead', 'CG', 'BFGS' or 'L-BFGS-B'.")
if (!(verbose %in% c(TRUE,FALSE))) stop("verbose must be TRUE or FALSE.")
}
check.factors <- function(long.df) {
if (is.factor(long.df$ROW)) {
message("Converting factor ROW to numeric.")
attributes(long.df)$ROWlevels <- levels(long.df$ROW)
long.df$ROW <- as.numeric(long.df$ROW)
}
if (is.factor(long.df$COL)) {
message("Converting factor COL to numeric.")
attributes(long.df)$COLlevels <- levels(long.df$COL)
long.df$COL <- as.numeric(long.df$COL)
}
# Check that all categorical covariates are factors
cov_idxs <- which(!(names(long.df) %in% c("Y","ROW","COL")))
for (j in cov_idxs) {
if (is.character(long.df[,j])) long.df[,j] <- factor(long.df[,j])
}
long.df
}
#' @importFrom stats model.matrix
check.formula <- function(formula, model, long.df, RG, CG) {
n <- max(long.df$ROW)
p <- max(long.df$COL)
q <- max(as.numeric(long.df$Y))
fo_terms <- terms(formula)
fo_vars <- as.character(unlist(as.list(attr(fo_terms,"variables"))))
fo_labels <- attr(fo_terms, "term.labels")
# Part A
# A.1 Check that Y is in the formula, and that it only occurs as-is, and only
# on the left-hand side of the formula
# OUTPUT: errors
if (is.na(match("Y",fo_vars))) stop("Y must appear in the formula as the response, and you cannot use a function of Y.")
if (attr(fo_terms,"response") != 1 || length(grep("Y", fo_labels)) > 0) {
stop("Y can only appear in the formula as the response.")
}
# A.2 Check that either RowClust or ColClust is in the formula, and check
# the presence of either or both against whether RG or CG is null
# OUTPUT: errors
rowc_grep <- length(grep("ROWCLUST",fo_vars))
colc_grep <- length(grep("COLCLUST",fo_vars))
if (rowc_grep == 0 && colc_grep == 0) {
stop("You must include ROWCLUST or COLCLUST in the formula.")
}
if (rowc_grep > 0 && is.null(RG)) stop("If you include ROWCLUST in the formula, you must also supply an integer value for nclus.row.")
if (colc_grep > 0 && is.null(CG)) stop("If you include COLCLUST in the formula, you must also supply an integer value for nclus.column.")
if (any(fo_labels == "ROWCLUST")) {
rowc_part <- "ROWCLUST"
}
if (any(fo_labels == "COLCLUST")) {
colc_part <- "COLCLUST"
}
# A.3 Check that there are no functions of RowClust or ColClust included in
# the formula (may have interaction terms, but no functions of them)
# OUTPUT: errors
if (rowc_grep > 0 && rowc_grep != sum(fo_vars == "ROWCLUST")) stop("You cannot use functions of ROWCLUST, only ROWCLUST as-is.")
if (colc_grep > 0 && colc_grep != sum(fo_vars == "COLCLUST")) stop("You cannot use functions of COLCLUST, only COLCLUST as-is.")
# A.4 Check that if the ROWCLUST:COLCLUST interaction term is included, that
# both or neither of the ROWCLUST, COLCLUST main effect terms are
# included
# OUTPUT: errors
if (any(fo_labels %in% c("ROWCLUST:COLCLUST","COLCLUST:ROWCLUST"))) {
if ((any(fo_labels == "ROWCLUST") && !any(fo_labels == "COLCLUST")) ||
(!any(fo_labels == "ROWCLUST") && any(fo_labels == "COLCLUST"))) {
stop("If including the interaction between row and column clustering, you must include both or neither of the main effects ROWCLUST and COLCLUST.")
}
}
# A.5 Check that if ROW or COL is in the formula, there are no functions of
# them, or any interaction terms involving them EXCEPT interactions
# between ROW and COLCLUST or between COL and ROWCLUST
# OUTPUT: errors
# We looked for ROWCLUST in attr(terms, "variables"), because ROWCLUST is
# allowed to have interaction terms, just not allowed to be included via a
# function of it. But for ROW we look in attr(terms, "term.labels"), because
# ROW can only be included in the terms as-is and on its own, not as part of
# any interactions
temp_labels <- gsub("ROWCLUST","ZZ",fo_labels)
row_grep <- length(grep("ROW",temp_labels))
if (row_grep > 0 && row_grep != sum(temp_labels %in% c("ROW","ROW:COLCLUST","COLCLUST:ROW"))) {
stop("You cannot use functions of ROW, and the only permitted interaction is with COLCLUST.")
}
if (row_grep > 0 && rowc_grep > 0) {
stop("You cannot include ROW as well as ROWCLUST.")
}
if (any(fo_labels %in% c("COLCLUST:ROW","ROW:COLCLUST")) &&
((any(fo_labels == "COLCLUST") && !any(fo_labels == "ROW")) ||
!any(fo_labels == "COLCLUST") && any(fo_labels == "ROW"))) {
stop("If including the interaction between column clusters and row effects, you must include both or neither of the main effects COLCLUST and ROW.")
}
if (any(temp_labels == "ROW")) {
row_part <- "ROW"
}
if (any(temp_labels %in% c("ROW:COLCLUST","COLCLUST:ROW"))) {
colc_row_part <- "COLCLUST:ROW"
}
temp_labels <- gsub("COLCLUST","ZZ",fo_labels)
col_grep <- length(grep("COL",temp_labels))
if (col_grep > 0 && col_grep != sum(temp_labels %in% c("COL","COL:ROWCLUST","ROWCLUST:COL"))) {
stop("You cannot use functions of COL, and the only permitted interaction is with ROWCLUST.")
}
if (col_grep > 0 && colc_grep > 0) {
stop("You cannot include COL as well as COLCLUST.")
}
if (any(fo_labels %in% c("ROWCLUST:COL","COL:ROWCLUST")) &&
((any(fo_labels == "ROWCLUST") && !any(fo_labels == "COL")) ||
!any(fo_labels == "ROWCLUST") && any(fo_labels == "COL"))) {
stop("If including the interaction between row clusters and column effects, you must include both or neither of the main effects ROWCLUST and COL.")
}
if (any(temp_labels == "COL")) {
col_part <- "COL"
}
if (any(temp_labels %in% c("COL:ROWCLUST","ROWCLUST:COL"))) {
rowc_col_part <- "ROWCLUST:COL"
}
# A.6 Check that there are no three-way or higher-order interactions
# involving ROWCLUST and COLCLUST
rowc_idxs <- grep("ROWCLUST",fo_labels)
colc_idxs <- grep("COLCLUST",fo_labels)
if (length(rowc_idxs) > 0 && length(colc_idxs) > 0) {
# Find terms that involve both ROWCLUST and COLCLUST, then exclude the
# term "ROWCLUST:COLCLUST" if it exists, because that interaction is not
# allowed.
rowc_colc_idxs <- match(rowc_idxs, colc_idxs)
rowc_colc_idxs <- rowc_colc_idxs[!is.na(rowc_colc_idxs)]
rowc_colc_labels <- fo_labels[colc_idxs[rowc_colc_idxs]]
rowc_colc_interaction_idxs <- match(c("ROWCLUST:COLCLUST","COLCLUST:ROWCLUST"), rowc_colc_labels)
if (any(!is.na(rowc_colc_interaction_idxs))) {
rowc_colc_interaction_idxs <- rowc_colc_interaction_idxs[!is.na(rowc_colc_interaction_idxs)]
rowc_colc_part <- rowc_colc_labels[rowc_colc_interaction_idxs]
rowc_colc_idxs <- rowc_colc_idxs[-rowc_colc_interaction_idxs]
}
if (length(rowc_colc_idxs) > 0) {
stop("If you include ROWCLUST and COLCLUST, you cannot include three-way or higher interactions that involve both ROWCLUST and COLCLUST.")
}
}
# A.7 Check that all other variables in the formula are in long.df, and in
# the process separate out the ROW/COL/ROWCLUST/COLCLUST parts, the ROWCLUST
# covariate parts, the COLCLUST covariate parts, and the pure covariate
# parts
# OUTPUT: errors
row_col_idxs <- which(fo_labels %in% c("ROW","COL","ROWCLUST","COLCLUST",
"ROWCLUST:COLCLUST","COLCLUST:ROWCLUST",
"ROWCLUST:COL","COL:ROWCLUST",
"COLCLUST:ROW","ROW:COLCLUST"))
if (length(row_col_idxs) > 0) {
row_col_part <- fo_labels[row_col_idxs]
non_row_col_part <- fo_labels[-row_col_idxs]
} else {
non_row_col_part <- fo_labels
}
rowc_idxs <- grep("ROWCLUST", non_row_col_part)
if (length(rowc_idxs) > 0) {
if (!any(fo_labels == "ROWCLUST")) {
stop("If you are including interactions between row clusters and covariates, you must include the main effect term for ROWCLUST.")
}
rowc_parts <- extract.covs("ROWCLUST", non_row_col_part[rowc_idxs], long.df)
rowc_cov_part <- rowc_parts$clust_cov_part
rowc_fo <- rowc_parts$clust_fo
rowc_mm <- rowc_parts$clust_mm
if (nrow(rowc_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
rowc_fo <- NULL
rowc_mm <- matrix(1)
}
colc_idxs <- grep("COLCLUST", non_row_col_part)
if (length(colc_idxs) > 0) {
if (!any(fo_labels == "COLCLUST")) {
stop("If you are including interactions between column clusters and covariates, you must include the main effect term for COLCLUST.")
}
colc_parts <- extract.covs("COLCLUST", non_row_col_part[colc_idxs], long.df)
colc_cov_part <- colc_parts$clust_cov_part
colc_fo <- colc_parts$clust_fo
colc_mm <- colc_parts$clust_mm
if (nrow(colc_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
colc_fo <- NULL
colc_mm <- matrix(1)
}
if (length(rowc_idxs) > 0 || length(colc_idxs) > 0) pure_cov_part <- non_row_col_part[-c(rowc_idxs, colc_idxs)]
else pure_cov_part <- non_row_col_part
if (length(pure_cov_part) > 0) {
cov_fo <- formula(paste("Y ~",paste(pure_cov_part, collapse="+")))
cov_tf <- terms(cov_fo)
# Now obtain model matrix, making sure to remove the intercept column AFTER
# constructing the matrix i.e. we want to have only the columns of the model
# matrix that would have been there alongside the intercept column
# In a regression model, if we take out the intercept term then there's
# scope for another independent column for the final category of any
# categorical variable, but we don't actually want to include that in our
# model matrix
cov_mm <- model.matrix(cov_tf, data=long.df)
# When dropping the intercept column, need to KEEP the matrix structure
cov_mm <- cov_mm[,-1, drop=FALSE]
if (nrow(cov_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
cov_fo <- NULL
cov_mm <- matrix(1)
}
# B.1 Construct list of params for ROW, COL, and pure RowClust and ColClust
# terms (i.e. all the remaining params apart from model-specific params
# like mu or mu_k, and phi_k)
# OUTPUT: params
param_lengths <- rep(0, 12)
names(param_lengths) <- c("mu","phi","rowc","col","rowc_col","rowc_cov","cov",
"colc","rowc_colc","row","colc_row","colc_cov")
param_lengths['mu'] <- q
if (model == "OSM") param_lengths['phi'] <- q
if (exists('rowc_part')) param_lengths['rowc'] <- RG
if (exists('colc_part')) param_lengths['colc'] <- CG
if (exists('rowc_colc_part')) param_lengths['rowc_colc'] <- RG*CG
if (exists('row_part')) param_lengths['row'] <- n
if (exists('colc_row_part')) param_lengths['colc_row'] <- CG*n
if (exists('col_part')) param_lengths['col'] <- p
if (exists('rowc_col_part')) param_lengths['rowc_col'] <- RG*p
## IMPORTANT: for the covariate sections, CANNOT just use the number of
## terms in the formula to indicate the number of covariate terms, because
## the formula only counts a categorical variable once, whereas the model
## matrix may have more columns than that, to accommodate the dummy
## variables for the levels of the categorical variable!
## Need to rely on model matrix instead!
if (exists('rowc_cov_part') && length(rowc_cov_part) > 0) param_lengths['rowc_cov'] <- dim(rowc_mm)[2]*RG
if (exists('colc_cov_part') && length(colc_cov_part) > 0) param_lengths['colc_cov'] <- dim(colc_mm)[2]*CG
if (exists('pure_cov_part') && length(pure_cov_part) > 0) param_lengths['cov'] <- dim(cov_mm)[2]
# Return model matrices
# Return list of params
list(param_lengths=param_lengths, rowc_fo=rowc_fo, rowc_mm=rowc_mm,
colc_fo=colc_fo, colc_mm=colc_mm, cov_fo=cov_fo, cov_mm=cov_mm)
}
## Conversion of the column clustering formula into row clustering format, in
## order for the actual clustering to be done via the row clustering functions
#' @importFrom stats terms formula
convert.model.row.to.column <- function(row.model_structure) {
pl <- row.model_structure$param_lengths
if (pl['colc'] > 0 && pl['rowc'] == 0) {
pl['rowc'] <- pl['colc']
pl['colc'] <- 0
}
if (pl['row'] > 0 && pl['col'] == 0) {
pl['col'] <- pl['row']
pl['row'] <- 0
}
if (pl['colc_row'] > 0 && pl['rowc_col'] == 0) {
pl['rowc_col'] <- pl['colc_row']
pl['colc_row'] <- 0
}
if (pl['colc_cov'] > 0 && pl['rowc_cov'] == 0) {
pl['rowc_cov'] <- pl['colc_cov']
pl['colc_cov'] <- 0
}
rowc_mm <- matrix(1)
rowc_fo <- NULL
colc_mm <- matrix(1)
colc_fo <- NULL
cov_mm <- row.model_structure$cov_mm
cov_fo <- row.model_structure$cov_fo
if (any(dim(row.model_structure$colc_mm) > 1) &&
all(dim(row.model_structure$rowc_mm) == 1)) {
rowc_mm <- row.model_structure$colc_mm
colc_mm <- matrix(1)
colc_terms <- terms(row.model_structure$colc_fo)
colc_labels <- attr(colc_terms, "term.labels")
rowc_fo <- formula(paste("Y ~",paste(colc_labels, collapse="+")))
colc_fo <- NULL
}
if (any(pl[c('colc','row','colc_row','colc_cov')] > 0) ||
any(dim(row.model_structure$rowc_mm) > 1)) stop("Error involving model structure.")
list(param_lengths=pl, rowc_fo=rowc_fo, rowc_mm=rowc_mm,
colc_fo=colc_fo, colc_mm=colc_mm, cov_fo=cov_fo, cov_mm=cov_mm)
}
#' @importFrom stats model.matrix
extract.covs <- function(clust_name, non_row_col_part, long.df) {
clust_cov_part <- non_row_col_part
# First, remove the clustering term from those parts, because we want to
# obtain the model matrix of the remaining parts of the terms
for (i in seq_along(clust_cov_part)) {
term <- clust_cov_part[i]
if (substr(term, 1, 9) == paste0(clust_name, ":")) {
clust_cov_part[i] <- substr(term,10,nchar(term))
} else {
clust_cov_part[i] <- sub(paste0(":",clust_name),"",term)
}
}
# Now obtain model matrix, making sure to remove the intercept column AFTER
# constructing the matrix i.e. we want to have only the columns of the model
# matrix that would have been there alongside the intercept column
# In a regression model, if we take out the intercept term then there's
# scope for another independent column for the final category of any
# categorical variable, but we don't actually want to include that in our
# model matrix
clust_fo <- formula(paste("Y ~",paste(clust_cov_part,collapse="+")))
clust_tf <- terms(clust_fo)
clust_mm <- model.matrix(clust_tf, data=long.df)
# When dropping the intercept column, need to KEEP the matrix structure
clust_mm <- clust_mm[,-1, drop=FALSE]
list(clust_cov_part=clust_cov_part, clust_fo=clust_fo, clust_mm=clust_mm)
}
Try the clustord package in your browser
Any scripts or data that you put into this service are public.
clustord documentation built on June 8, 2025, 1:03 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.
Defines functions extract.covs convert.model.row.to.column check.formula check.factors validate.inputs
#' @importFrom methods is
validate.inputs <- function(formula, model,
nclus.row=NULL,nclus.column=NULL,
long.df,
initvect=NULL,
pi.init=NULL, kappa.init=NULL,
EM.control=default.EM.control(),
optim.method="L-BFGS-B",
constraint_sum_zero=TRUE,
start_from_simple_model=TRUE,
parallel_starts=FALSE,
nstarts=5,
verbose=TRUE) {
## Note the double-& and double-| which stops the later parts being checked
## if the earlier parts are false
if (!is(formula, "formula")) stop("formula must be a valid formula.")
## Check that model is valid
if (!is.character(model) || !is.vector(model) || length(model) != 1) stop("model must be a string, 'OSM' or 'POM' or 'Binary'.")
if (!(model %in% c("OSM","POM","Binary"))) stop("model must be either 'OSM' or POM' for the ordered stereotype and proportional odds models, or 'Binary' for the binary model.")
## Check that clustering settings are valid (later, in check.formula, will
## make sure user has provide nclus.row or nclus.column or both, depending
## on formula supplied)
if (!is.null(nclus.row)) {
if (!is.vector(nclus.row) || length(nclus.row) != 1 || nclus.row <= 1 ||
nclus.row %% 1 != 0 || is.na(nclus.row)) {
stop("nclus.row must be an integer, from 2 to the number of rows/observations in the data.")
}
}
if (!is.null(nclus.column)) {
if (!is.vector(nclus.column) || length(nclus.column) != 1 ||
nclus.column <= 1 || nclus.column %% 1 != 0 || is.na(nclus.column)) {
stop("nclus.column must be an integer, from 2 to the number of columns/questions in the data.")
}
}
if (is.null(long.df)) stop("long.df cannot be null.")
if (!is.data.frame(long.df)) stop("long.df must be a data frame.")
if (length(long.df) < 3) stop("long.df must have at least 3 columns, Y and ROW and COL.")
names.df <- names(long.df)
# Convert the Y, ROW and COL variables to upper case if they are in the df but
# in the wrong case
if ("y" %in% names.df) names.df[which(names.df == "y")] <- "Y"
if ("ROW" %in% toupper(names.df)) names.df[which(toupper(names.df) == "ROW")] <- "ROW"
if ("COL" %in% toupper(names.df)) names.df[which(toupper(names.df) == "COL")] <- "COL"
if (!("Y" %in% names.df)) stop("long.df must have a column named 'Y' which contains the response values.")
if (!("ROW" %in% names.df)) stop("long.df must have a column named 'ROW' which indicates what observation (row in the data matrix) each value of Y corresponds to.")
if (!("COL" %in% names.df)) stop("long.df must have a column named 'COL' which indicates what variable (column in the data matrix) each value of Y corresponds to.")
if (length(which(names.df == "Y")) > 1) stop("long.df should only have one column named 'Y'.")
if (length(which(names.df == "ROW")) > 1) stop("long.df should only have one column named 'ROW'.")
if (length(which(names.df == "COL")) > 1) stop("long.df should only have one column named 'COL'.")
if (!is.factor(long.df$Y)) stop("long.df$Y must be a factor.")
if (any(is.na(long.df$Y))) stop("long.df$Y has missing values (NA). Please delete these rows and try again.")
if (is.list(long.df$Y) || any(sapply(long.df$Y,is.list)) ||
any(sapply(long.df$Y,is.infinite))) stop("long.df$Y is a list, or has list elements or infinite elements. long.df$Y should be a factor with q levels.")
if (model == "Binary" && length(unique(long.df$Y)) > 2) stop("For the Binary model, long.df$Y should only have 2 possible values.")
if (!is.factor(long.df$ROW) &&
(is.list(long.df$ROW) || any(sapply(long.df$ROW,is.list)) || any(is.na(long.df$ROW)) ||
any(sapply(long.df$ROW,is.infinite)) || any(long.df$ROW %% 1 != 0) ||
any(long.df$ROW < 1) || all(long.df$ROW > 1))) stop("long.df$ROW must be a factor or integers from 1 to the number of observations, i.e. the number of rows in the original data matrix.")
if (!is.factor(long.df$COL) &&
(is.list(long.df$COL) || any(sapply(long.df$COL,is.list)) || any(is.na(long.df$COL)) ||
any(sapply(long.df$COL,is.infinite)) || any(long.df$COL %% 1 != 0) ||
any(long.df$COL < 1) || all(long.df$COL > 1))) stop("long.df$COL must be a factor or integers from 1 to the number of variables, i.e. the number of columns in the original data matrix.")
if (any(table(long.df[,c("ROW","COL")]) > 1)) stop("Each element from the original data matrix must correspond to no more than 1 row in long.df.")
cov_idxs <- which(!(names(long.df) %in% c("Y","ROW","COL")))
for (j in cov_idxs) {
col <- long.df[,j]
non_na_col <- col[!is.na(col)]
if (length(unique(non_na_col)) == 1) stop(paste("Covariate",names(long.df)[j],"only takes one non-missing value for all entries of the data matrix. Please remove this covariate before continuing."))
}
if (!is.null(nclus.row) && nclus.row >= max(as.numeric(long.df$ROW))) stop("nclus.row must be smaller than the maximum value of long.df$ROW.")
if (!is.null(nclus.column) && nclus.column >= max(as.numeric(long.df$COL))) stop("nclus.column must be smaller than the maximum value of long.df$COL.")
if (!is.null(initvect)) {
if (!is.vector(initvect) || !is.numeric(initvect) || any(is.na(initvect)) ||
any(is.infinite(initvect))) stop("If supplied, initvect must be a numeric vector with finite values.")
}
if (!is.null(pi.init)) {
if (!is.vector(pi.init) || !is.numeric(pi.init) || any(is.na(pi.init)) ||
any(pi.init < 0) || any(pi.init > 1)) stop("If supplied, pi.init must be a vector of numbers between 0 and 1.")
if (length(pi.init) != nclus.row || abs(sum(pi.init) - 1) > 1e-12) stop("pi.init must be the same length as the number of row clusters, and must add up to 1")
}
if (!is.null(kappa.init)) {
if (!is.vector(kappa.init) || !is.numeric(kappa.init) || any(is.na(kappa.init)) ||
any(kappa.init < 0) | any(kappa.init > 1)) stop("If supplied, kappa.init must be a vector of numbers between 0 and 1.")
if (length(kappa.init) != nclus.column || abs(sum(kappa.init) - 1) > 1e-12) stop("kappa.init must be the same length as the number of column clusters, and must add up to 1")
}
if (!is.logical(constraint_sum_zero) || !is.vector(constraint_sum_zero) ||
length(constraint_sum_zero) != 1 || is.na(constraint_sum_zero)) stop("constraint_sum_zero must be TRUE or FALSE.")
if (!is.logical(start_from_simple_model) || !is.vector(start_from_simple_model) ||
length(start_from_simple_model) != 1 || is.na(start_from_simple_model)) stop("start_from_simple_model must be TRUE or FALSE.")
if (!is.null(nstarts)) {
if (!is.vector(nstarts) || !is.numeric(nstarts) || length(nstarts) != 1 ||
nstarts < 0 || nstarts %% 1 != 0) stop("If supplied, nstarts must be a positive integer.")
}
if (!(parallel_starts %in% c(TRUE,FALSE))) {
stop("parallel_starts must be TRUE or FALSE.")
}
if (!is.null(EM.control) & (!is.list(EM.control) || length(EM.control) == 0 || length(EM.control) > 9 ||
!all(names(EM.control) %in% c("EMcycles","EMlikelihoodtol","EMparamtol",
"paramstopping","startEMcycles","keepallparams",
"epsilon", "rerunestepbeforelli",
"uselatestlli")))) {
stop("If supplied, EM.control must be a list of control parameters for the EM algorithm. Please see the manual for more info.")
}
if (is.null(optim.method) || !is.character(optim.method) || !is.vector(optim.method) ||
length(optim.method) != 1 || !(optim.method %in% c("Nelder-Mead","BFGS","CG","L-BFGS-B"))) stop("If supplied, optim.method must be one of the valid methods for optim, 'Nelder-Mead', 'CG', 'BFGS' or 'L-BFGS-B'.")
if (!(verbose %in% c(TRUE,FALSE))) stop("verbose must be TRUE or FALSE.")
}
check.factors <- function(long.df) {
if (is.factor(long.df$ROW)) {
message("Converting factor ROW to numeric.")
attributes(long.df)$ROWlevels <- levels(long.df$ROW)
long.df$ROW <- as.numeric(long.df$ROW)
}
if (is.factor(long.df$COL)) {
message("Converting factor COL to numeric.")
attributes(long.df)$COLlevels <- levels(long.df$COL)
long.df$COL <- as.numeric(long.df$COL)
}
# Check that all categorical covariates are factors
cov_idxs <- which(!(names(long.df) %in% c("Y","ROW","COL")))
for (j in cov_idxs) {
if (is.character(long.df[,j])) long.df[,j] <- factor(long.df[,j])
}
long.df
}
#' @importFrom stats model.matrix
check.formula <- function(formula, model, long.df, RG, CG) {
n <- max(long.df$ROW)
p <- max(long.df$COL)
q <- max(as.numeric(long.df$Y))
fo_terms <- terms(formula)
fo_vars <- as.character(unlist(as.list(attr(fo_terms,"variables"))))
fo_labels <- attr(fo_terms, "term.labels")
# Part A
# A.1 Check that Y is in the formula, and that it only occurs as-is, and only
# on the left-hand side of the formula
# OUTPUT: errors
if (is.na(match("Y",fo_vars))) stop("Y must appear in the formula as the response, and you cannot use a function of Y.")
if (attr(fo_terms,"response") != 1 || length(grep("Y", fo_labels)) > 0) {
stop("Y can only appear in the formula as the response.")
}
# A.2 Check that either RowClust or ColClust is in the formula, and check
# the presence of either or both against whether RG or CG is null
# OUTPUT: errors
rowc_grep <- length(grep("ROWCLUST",fo_vars))
colc_grep <- length(grep("COLCLUST",fo_vars))
if (rowc_grep == 0 && colc_grep == 0) {
stop("You must include ROWCLUST or COLCLUST in the formula.")
}
if (rowc_grep > 0 && is.null(RG)) stop("If you include ROWCLUST in the formula, you must also supply an integer value for nclus.row.")
if (colc_grep > 0 && is.null(CG)) stop("If you include COLCLUST in the formula, you must also supply an integer value for nclus.column.")
if (any(fo_labels == "ROWCLUST")) {
rowc_part <- "ROWCLUST"
}
if (any(fo_labels == "COLCLUST")) {
colc_part <- "COLCLUST"
}
# A.3 Check that there are no functions of RowClust or ColClust included in
# the formula (may have interaction terms, but no functions of them)
# OUTPUT: errors
if (rowc_grep > 0 && rowc_grep != sum(fo_vars == "ROWCLUST")) stop("You cannot use functions of ROWCLUST, only ROWCLUST as-is.")
if (colc_grep > 0 && colc_grep != sum(fo_vars == "COLCLUST")) stop("You cannot use functions of COLCLUST, only COLCLUST as-is.")
# A.4 Check that if the ROWCLUST:COLCLUST interaction term is included, that
# both or neither of the ROWCLUST, COLCLUST main effect terms are
# included
# OUTPUT: errors
if (any(fo_labels %in% c("ROWCLUST:COLCLUST","COLCLUST:ROWCLUST"))) {
if ((any(fo_labels == "ROWCLUST") && !any(fo_labels == "COLCLUST")) ||
(!any(fo_labels == "ROWCLUST") && any(fo_labels == "COLCLUST"))) {
stop("If including the interaction between row and column clustering, you must include both or neither of the main effects ROWCLUST and COLCLUST.")
}
}
# A.5 Check that if ROW or COL is in the formula, there are no functions of
# them, or any interaction terms involving them EXCEPT interactions
# between ROW and COLCLUST or between COL and ROWCLUST
# OUTPUT: errors
# We looked for ROWCLUST in attr(terms, "variables"), because ROWCLUST is
# allowed to have interaction terms, just not allowed to be included via a
# function of it. But for ROW we look in attr(terms, "term.labels"), because
# ROW can only be included in the terms as-is and on its own, not as part of
# any interactions
temp_labels <- gsub("ROWCLUST","ZZ",fo_labels)
row_grep <- length(grep("ROW",temp_labels))
if (row_grep > 0 && row_grep != sum(temp_labels %in% c("ROW","ROW:COLCLUST","COLCLUST:ROW"))) {
stop("You cannot use functions of ROW, and the only permitted interaction is with COLCLUST.")
}
if (row_grep > 0 && rowc_grep > 0) {
stop("You cannot include ROW as well as ROWCLUST.")
}
if (any(fo_labels %in% c("COLCLUST:ROW","ROW:COLCLUST")) &&
((any(fo_labels == "COLCLUST") && !any(fo_labels == "ROW")) ||
!any(fo_labels == "COLCLUST") && any(fo_labels == "ROW"))) {
stop("If including the interaction between column clusters and row effects, you must include both or neither of the main effects COLCLUST and ROW.")
}
if (any(temp_labels == "ROW")) {
row_part <- "ROW"
}
if (any(temp_labels %in% c("ROW:COLCLUST","COLCLUST:ROW"))) {
colc_row_part <- "COLCLUST:ROW"
}
temp_labels <- gsub("COLCLUST","ZZ",fo_labels)
col_grep <- length(grep("COL",temp_labels))
if (col_grep > 0 && col_grep != sum(temp_labels %in% c("COL","COL:ROWCLUST","ROWCLUST:COL"))) {
stop("You cannot use functions of COL, and the only permitted interaction is with ROWCLUST.")
}
if (col_grep > 0 && colc_grep > 0) {
stop("You cannot include COL as well as COLCLUST.")
}
if (any(fo_labels %in% c("ROWCLUST:COL","COL:ROWCLUST")) &&
((any(fo_labels == "ROWCLUST") && !any(fo_labels == "COL")) ||
!any(fo_labels == "ROWCLUST") && any(fo_labels == "COL"))) {
stop("If including the interaction between row clusters and column effects, you must include both or neither of the main effects ROWCLUST and COL.")
}
if (any(temp_labels == "COL")) {
col_part <- "COL"
}
if (any(temp_labels %in% c("COL:ROWCLUST","ROWCLUST:COL"))) {
rowc_col_part <- "ROWCLUST:COL"
}
# A.6 Check that there are no three-way or higher-order interactions
# involving ROWCLUST and COLCLUST
rowc_idxs <- grep("ROWCLUST",fo_labels)
colc_idxs <- grep("COLCLUST",fo_labels)
if (length(rowc_idxs) > 0 && length(colc_idxs) > 0) {
# Find terms that involve both ROWCLUST and COLCLUST, then exclude the
# term "ROWCLUST:COLCLUST" if it exists, because that interaction is not
# allowed.
rowc_colc_idxs <- match(rowc_idxs, colc_idxs)
rowc_colc_idxs <- rowc_colc_idxs[!is.na(rowc_colc_idxs)]
rowc_colc_labels <- fo_labels[colc_idxs[rowc_colc_idxs]]
rowc_colc_interaction_idxs <- match(c("ROWCLUST:COLCLUST","COLCLUST:ROWCLUST"), rowc_colc_labels)
if (any(!is.na(rowc_colc_interaction_idxs))) {
rowc_colc_interaction_idxs <- rowc_colc_interaction_idxs[!is.na(rowc_colc_interaction_idxs)]
rowc_colc_part <- rowc_colc_labels[rowc_colc_interaction_idxs]
rowc_colc_idxs <- rowc_colc_idxs[-rowc_colc_interaction_idxs]
}
if (length(rowc_colc_idxs) > 0) {
stop("If you include ROWCLUST and COLCLUST, you cannot include three-way or higher interactions that involve both ROWCLUST and COLCLUST.")
}
}
# A.7 Check that all other variables in the formula are in long.df, and in
# the process separate out the ROW/COL/ROWCLUST/COLCLUST parts, the ROWCLUST
# covariate parts, the COLCLUST covariate parts, and the pure covariate
# parts
# OUTPUT: errors
row_col_idxs <- which(fo_labels %in% c("ROW","COL","ROWCLUST","COLCLUST",
"ROWCLUST:COLCLUST","COLCLUST:ROWCLUST",
"ROWCLUST:COL","COL:ROWCLUST",
"COLCLUST:ROW","ROW:COLCLUST"))
if (length(row_col_idxs) > 0) {
row_col_part <- fo_labels[row_col_idxs]
non_row_col_part <- fo_labels[-row_col_idxs]
} else {
non_row_col_part <- fo_labels
}
rowc_idxs <- grep("ROWCLUST", non_row_col_part)
if (length(rowc_idxs) > 0) {
if (!any(fo_labels == "ROWCLUST")) {
stop("If you are including interactions between row clusters and covariates, you must include the main effect term for ROWCLUST.")
}
rowc_parts <- extract.covs("ROWCLUST", non_row_col_part[rowc_idxs], long.df)
rowc_cov_part <- rowc_parts$clust_cov_part
rowc_fo <- rowc_parts$clust_fo
rowc_mm <- rowc_parts$clust_mm
if (nrow(rowc_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
rowc_fo <- NULL
rowc_mm <- matrix(1)
}
colc_idxs <- grep("COLCLUST", non_row_col_part)
if (length(colc_idxs) > 0) {
if (!any(fo_labels == "COLCLUST")) {
stop("If you are including interactions between column clusters and covariates, you must include the main effect term for COLCLUST.")
}
colc_parts <- extract.covs("COLCLUST", non_row_col_part[colc_idxs], long.df)
colc_cov_part <- colc_parts$clust_cov_part
colc_fo <- colc_parts$clust_fo
colc_mm <- colc_parts$clust_mm
if (nrow(colc_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
colc_fo <- NULL
colc_mm <- matrix(1)
}
if (length(rowc_idxs) > 0 || length(colc_idxs) > 0) pure_cov_part <- non_row_col_part[-c(rowc_idxs, colc_idxs)]
else pure_cov_part <- non_row_col_part
if (length(pure_cov_part) > 0) {
cov_fo <- formula(paste("Y ~",paste(pure_cov_part, collapse="+")))
cov_tf <- terms(cov_fo)
# Now obtain model matrix, making sure to remove the intercept column AFTER
# constructing the matrix i.e. we want to have only the columns of the model
# matrix that would have been there alongside the intercept column
# In a regression model, if we take out the intercept term then there's
# scope for another independent column for the final category of any
# categorical variable, but we don't actually want to include that in our
# model matrix
cov_mm <- model.matrix(cov_tf, data=long.df)
# When dropping the intercept column, need to KEEP the matrix structure
cov_mm <- cov_mm[,-1, drop=FALSE]
if (nrow(cov_mm) < nrow(long.df)) stop("NA or NaN rows have been dropped from model matrix, possibly after calculating covariate terms derived from raw covariates. clustord cannot yet handle missing data in covariates. Please check your formula and covariate values and try again.")
} else {
cov_fo <- NULL
cov_mm <- matrix(1)
}
# B.1 Construct list of params for ROW, COL, and pure RowClust and ColClust
# terms (i.e. all the remaining params apart from model-specific params
# like mu or mu_k, and phi_k)
# OUTPUT: params
param_lengths <- rep(0, 12)
names(param_lengths) <- c("mu","phi","rowc","col","rowc_col","rowc_cov","cov",
"colc","rowc_colc","row","colc_row","colc_cov")
param_lengths['mu'] <- q
if (model == "OSM") param_lengths['phi'] <- q
if (exists('rowc_part')) param_lengths['rowc'] <- RG
if (exists('colc_part')) param_lengths['colc'] <- CG
if (exists('rowc_colc_part')) param_lengths['rowc_colc'] <- RG*CG
if (exists('row_part')) param_lengths['row'] <- n
if (exists('colc_row_part')) param_lengths['colc_row'] <- CG*n
if (exists('col_part')) param_lengths['col'] <- p
if (exists('rowc_col_part')) param_lengths['rowc_col'] <- RG*p
## IMPORTANT: for the covariate sections, CANNOT just use the number of
## terms in the formula to indicate the number of covariate terms, because
## the formula only counts a categorical variable once, whereas the model
## matrix may have more columns than that, to accommodate the dummy
## variables for the levels of the categorical variable!
## Need to rely on model matrix instead!
if (exists('rowc_cov_part') && length(rowc_cov_part) > 0) param_lengths['rowc_cov'] <- dim(rowc_mm)[2]*RG
if (exists('colc_cov_part') && length(colc_cov_part) > 0) param_lengths['colc_cov'] <- dim(colc_mm)[2]*CG
if (exists('pure_cov_part') && length(pure_cov_part) > 0) param_lengths['cov'] <- dim(cov_mm)[2]
# Return model matrices
# Return list of params
list(param_lengths=param_lengths, rowc_fo=rowc_fo, rowc_mm=rowc_mm,
colc_fo=colc_fo, colc_mm=colc_mm, cov_fo=cov_fo, cov_mm=cov_mm)
}
## Conversion of the column clustering formula into row clustering format, in
## order for the actual clustering to be done via the row clustering functions
#' @importFrom stats terms formula
convert.model.row.to.column <- function(row.model_structure) {
pl <- row.model_structure$param_lengths
if (pl['colc'] > 0 && pl['rowc'] == 0) {
pl['rowc'] <- pl['colc']
pl['colc'] <- 0
}
if (pl['row'] > 0 && pl['col'] == 0) {
pl['col'] <- pl['row']
pl['row'] <- 0
}
if (pl['colc_row'] > 0 && pl['rowc_col'] == 0) {
pl['rowc_col'] <- pl['colc_row']
pl['colc_row'] <- 0
}
if (pl['colc_cov'] > 0 && pl['rowc_cov'] == 0) {
pl['rowc_cov'] <- pl['colc_cov']
pl['colc_cov'] <- 0
}
rowc_mm <- matrix(1)
rowc_fo <- NULL
colc_mm <- matrix(1)
colc_fo <- NULL
cov_mm <- row.model_structure$cov_mm
cov_fo <- row.model_structure$cov_fo
if (any(dim(row.model_structure$colc_mm) > 1) &&
all(dim(row.model_structure$rowc_mm) == 1)) {
rowc_mm <- row.model_structure$colc_mm
colc_mm <- matrix(1)
colc_terms <- terms(row.model_structure$colc_fo)
colc_labels <- attr(colc_terms, "term.labels")
rowc_fo <- formula(paste("Y ~",paste(colc_labels, collapse="+")))
colc_fo <- NULL
}
if (any(pl[c('colc','row','colc_row','colc_cov')] > 0) ||
any(dim(row.model_structure$rowc_mm) > 1)) stop("Error involving model structure.")
list(param_lengths=pl, rowc_fo=rowc_fo, rowc_mm=rowc_mm,
colc_fo=colc_fo, colc_mm=colc_mm, cov_fo=cov_fo, cov_mm=cov_mm)
}
#' @importFrom stats model.matrix
extract.covs <- function(clust_name, non_row_col_part, long.df) {
clust_cov_part <- non_row_col_part
# First, remove the clustering term from those parts, because we want to
# obtain the model matrix of the remaining parts of the terms
for (i in seq_along(clust_cov_part)) {
term <- clust_cov_part[i]
if (substr(term, 1, 9) == paste0(clust_name, ":")) {
clust_cov_part[i] <- substr(term,10,nchar(term))
} else {
clust_cov_part[i] <- sub(paste0(":",clust_name),"",term)
}
}
# Now obtain model matrix, making sure to remove the intercept column AFTER
# constructing the matrix i.e. we want to have only the columns of the model
# matrix that would have been there alongside the intercept column
# In a regression model, if we take out the intercept term then there's
# scope for another independent column for the final category of any
# categorical variable, but we don't actually want to include that in our
# model matrix
clust_fo <- formula(paste("Y ~",paste(clust_cov_part,collapse="+")))
clust_tf <- terms(clust_fo)
clust_mm <- model.matrix(clust_tf, data=long.df)
# When dropping the intercept column, need to KEEP the matrix structure
clust_mm <- clust_mm[,-1, drop=FALSE]
list(clust_cov_part=clust_cov_part, clust_fo=clust_fo, clust_mm=clust_mm)
}
Try the clustord 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.