diff options
Diffstat (limited to '')
6 files changed, 0 insertions, 330 deletions
diff --git a/admin/survey/modules/mod_360_1KA/results/.gitignore b/admin/survey/modules/mod_360_1KA/results/.gitignore deleted file mode 100644 index a3a0c8b..0000000 --- a/admin/survey/modules/mod_360_1KA/results/.gitignore +++ /dev/null @@ -1,2 +0,0 @@ -*
-!.gitignore
\ No newline at end of file diff --git a/admin/survey/modules/mod_kakovost/R/calc.usability.R b/admin/survey/modules/mod_kakovost/R/calc.usability.R deleted file mode 100644 index 4e4bb0c..0000000 --- a/admin/survey/modules/mod_kakovost/R/calc.usability.R +++ /dev/null @@ -1,72 +0,0 @@ -calc.usability <- function(m.all, return.type){
- # return.type:
- # 1: return only absolute
- # 2: return only %
- # 3: return both (even rows: absolute, odd rows: %)
-
- ## calculations
- m.all[, Prekinitve:=v3]
- m.all[, Neodgovori:=v1]
- m.all[, Nevsebinski:=v96+v97+v98+v99]
- m.all[, Izpostavljen:=allqs-(v2+v3+v4+v5)]
- setnames(m.all, "va", "Veljavni")
-
- m.all[, UNL:=Neodgovori/Izpostavljen]
- m.all[is.na(UNL)==T, UNL:=0]
- m.all[, UML:=(v3/allqs)+(1-(v3/allqs))*UNL]
- m.all[, UCL:=1-UML]
- m.all[, UIL:=v2/(v2+Izpostavljen)]
- m.all[is.na(UIL)==T, UIL:=0]
- m.all[, UAQ:=v4/allqs]
-
- m.all[, Uporabnost:=1-UML]
-
- #tidy up
- setcolorder(m.all, c("recnum", "allqs", "Veljavni", "Nevsebinski", "Neodgovori",
- "Izpostavljen", "Prekinitve", "Uporabnost",
- "v1", "v2", "v3", "v4", "v5", "v96", "v97", "v98", "v99",
- "UNL", "UML", "UCL", "UIL", "UAQ"))
-
- if(return.type==1){
- return(m.all)
- }else{
- m.all.p <- copy(m.all)
-
- m.all.p[, (c("Veljavni", "Nevsebinski", "Neodgovori")) := lapply(.SD, "/", m.all.p$Izpostavljen), .SDcols=c("Veljavni", "Nevsebinski", "Neodgovori")]
- m.all.p[, (c("Prekinitve", "v1", "v2", "v3", "v4", "v5", "v96", "v97", "v98", "v99")) := lapply(.SD, "/", m.all.p$allqs), .SDcols=c("Prekinitve", "v1", "v2", "v3", "v4", "v5", "v96", "v97", "v98", "v99")]
- m.all.p[, Izpostavljen:=1]
-
- if(return.type==2){
- return(m.all.p)
- }else{
- m.all[, Uporabnost:=Veljavni]
- m.all[, c("UNL", "UML", "UCL", "UIL", "UAQ"):=NA]
- m.all <- m.all[, lapply(.SD, as.character)]
-
- m.all.p[, allqs:=NA]
- m.all.p[, allqs:=as.character(allqs)]
-
- change.cols <- c("Veljavni", "Nevsebinski", "Neodgovori", "Izpostavljen", "Prekinitve", "Uporabnost",
- "v1", "v2", "v3", "v4", "v5", "v96", "v97", "v98", "v99",
- "UNL", "UML", "UCL", "UIL", "UAQ")
- m.all.p[, (change.cols):=lapply(.SD, function(x){paste0(round(x*100, 0), "%")}), .SD=change.cols]
-
- m.1ka <- data.table(matrix("", nrow=nrow(m.all)*2, ncol=ncol(m.all)))
-
- a.rows <- as.integer(seq(1, nrow(m.1ka), by=2))
- p.rows <- as.integer(seq(2, nrow(m.1ka), by=2))
-
- set(m.1ka, a.rows, 1:ncol(m.1ka), value=m.all)
- suppressWarnings(set(m.1ka, p.rows, 1:ncol(m.1ka), value=m.all.p))
-
- setnames(m.1ka, colnames(m.all))
- m.1ka[, Status:=NA_character_]
- setcolorder(m.1ka, c("recnum", "allqs", "Veljavni", "Nevsebinski", "Neodgovori",
- "Izpostavljen", "Prekinitve", "Uporabnost", "Status",
- "v1", "v2", "v3", "v4", "v5", "v96", "v97", "v98", "v99",
- "UNL", "UML", "UCL", "UIL", "UAQ"))
-
- return(m.1ka)
- }
- }
-}
\ No newline at end of file diff --git a/admin/survey/modules/mod_kakovost/R/gen.survey.str.R b/admin/survey/modules/mod_kakovost/R/gen.survey.str.R deleted file mode 100644 index 32e1f57..0000000 --- a/admin/survey/modules/mod_kakovost/R/gen.survey.str.R +++ /dev/null @@ -1,71 +0,0 @@ -gen.survey.str <- function(colnames.dsa, questions.file, items.file){
- #import questions file
- questions <- fread(questions.file, skip=1, header=F,
- select=c(2, 5, 6, 8, 9, 10),
- col.names=c("question.id", "variable", "tip", "size", "visible", "params"))
-
- #create variable list from survey data file
- #remove "recnum" and "_text" fields
- var.data <- colnames.dsa[sapply(colnames.dsa, function(x){substr(x, nchar(x)-4, nchar(x))})!="_text"]
-
- #create variable list from questions file
- var.questions <- questions$variable
-
- #generate data.table from var.data list
- survey.str <- data.table(variable = var.data)
-
- setkey(questions, "variable")
- setkey(survey.str, "variable")
-
- #if all var.data in var.questions, do the simple merge and return file
- if(all(var.data %in% var.questions)){
- survey.str <- questions[survey.str,]
- return(survey.str)
- }else{ #if not, import items file and do additional merge with it...
- #import items file
- items <- fread(items.file, skip=1, header=F,
- select=c(2, 3, 4),
- col.names=c("question.id", "item.id", "variable"))
-
- setkey(items, "question.id")
- setkey(questions, "question.id")
-
- #bind variables from questions and items (for the later, only take instances with no match in the questions file...)
- survey.str.qi <- rbindlist(list(questions[var.questions %in% var.data,],
- items[questions[!(var.questions %in% var.data), -"variable", with=F], nomatch=0L]),
- fill=T)
-
- #merge questions+items with survey data...
- setkey(survey.str.qi, "variable")
- setkey(survey.str, "variable")
- survey.str <- survey.str.qi[survey.str,]
-
- #if all var.data is now matched, return the survey.str
- if(!(any(is.na(survey.str)))){
- return(survey.str)
- }else{ #if not, do additional merging...
- #create index of all NA instaces from survey.str...
- index <- apply(cbind(survey.str[, is.na(tip)],
- (sapply(survey.str[, variable], function(x){
- substr(x, 1, regexpr("\\_[^\\_]*$", x)-1)
- }) %in% survey.str.qi$variable)
- ),
- 1, all)
-
- #... using regex to find matches among unmatched instances from survey.str.qi
- add <- merge(survey.str[index, list(variable, substr(variable, 1, regexpr("\\_[^\\_]*$", variable)-1))],
- survey.str.qi[!(variable %in% survey.str$variable),],
- by.x="V2", by.y="variable", all.y=F)[, list(question.id, item.id, tip, visible, size, params)]
-
- #update survey.str with new values
- survey.str[index, c("question.id", "item.id", "tip", "visible", "size", "params") := as.list(add)]
-
- #if there is no NAs left, return survey.str, else return msg
- if(!(any(is.na(survey.str$tip)))){
- return(survey.str)
- }else{
- return(paste("No match found for: ", survey.str[is.na(tip), variable]))
- }
- }
- }
-}
diff --git a/admin/survey/modules/mod_kakovost/R/gen.usability.matrix.R b/admin/survey/modules/mod_kakovost/R/gen.usability.matrix.R deleted file mode 100644 index a2b1465..0000000 --- a/admin/survey/modules/mod_kakovost/R/gen.usability.matrix.R +++ /dev/null @@ -1,181 +0,0 @@ -gen.usability.matrix <- function(dsa, survey.str){
- #define special values to detect
- #order of this values is important:
- # in case of conflicts @ chk.t types of questions the order sets the priporty of which values to keep
- special.v <- c(-1, -3, -5, -96, -97, -98, -99, -4, -2)
-
- #define which variables belong to checkbox-like* questions
- #(* i.e.: check for special values @ ANY variable per question/item ID)
- # 2: normal checkbox
- # 16: multicheckbox
- # 17: ranking
- chkbox.t <- c(2, 16, 17)
-
- ##all other variables belong to normal** questions
- #(** i.e.: check for special values @ each variable per question/item ID)
- #if there are no normal questions, create 0 matrix, otherwise...
- if(nrow(survey.str[!(tip %in% chkbox.t),])==0){
- m.n <- matrix(0, nrow = nrow(dsa), ncol=length(special.v)+1)
- }else{
- #create list of all normal questions
- c.n <- colnames(dsa)[which(colnames(dsa) %in% survey.str[!(tip %in% chkbox.t), variable])]
-
- #...count all non-special values for each variable
- #... + count each special value for each variable
- m.n <- cbind(rowSums(sapply(dsa[, c.n, with=FALSE], function(x){!(x %in% special.v)})),
- sapply(special.v, function(x){as.integer(rowSums(dsa[, c.n, with=FALSE]==x, na.rm=TRUE))}))
- }
-
- ##procedure for tip:2
- #only run if there is an at least one tip:2 variable
- if(survey.str[, any(tip==2)]){
- #get list of all unique tip:2 question ids
- q.2 <- unique(survey.str[tip==2, question.id])
- #get list of all corresponding variables for each q.2 id
- c.2 <- lapply(q.2, function(x){colnames(dsa)[which(colnames(dsa) %in% survey.str[question.id==x & tip==2, variable])]})
-
- #(do this for each instance in c.2):
- #for each set of variables:
- # check if any variable contains at least one non-special value
- # + (for each special value) check if any variable contains at least special value
- m.2 <- lapply(c.2, function(x){
- cbind(apply(dsa[, x, with=FALSE], 1, function(q){any(!(q %in% special.v))}),
- sapply(special.v, function(y){
- apply(dsa[, x, with=FALSE], 1, function(q){any(q==y)})
- })
- )
- })
-
- # (do this for each instance in c.2)
- # if multiple special values per respondent exist, keep only the first one
- m.2 <- lapply(m.2, function(x){
- if(any(rowSums(x)>1)){
- p <- x[rowSums(x)>1,]
- for(i in 1:nrow(p)){
- a <- p[i,]
- f <- TRUE
- for(j in 1:length(a)){
- print(j)
- if(a[j] & f){
- f <- FALSE
- }else if(a[j] & !f){
- a[j] <- FALSE
- }
- }
- p[i,] <- a
- }
- x[rowSums(x)>1,] <- p
- }else{x}
- })
-
-
- #add to m.n
- m.n <- m.n + Reduce('+', m.2)
- }
-
- ##procedure for tip:16
- #only run if there is an at least one tip:16 variable
- if(survey.str[, any(tip==16)]){
- #get list of all unique tip:16 item ids
- q.16 <- unique(survey.str[tip==16, item.id])
-
- #get list of all corresponding variables for each q.16 id
- c.16 <- lapply(q.16, function(x){colnames(dsa)[which(colnames(dsa) %in% survey.str[item.id==x & tip==16, variable])]})
- #(do this for each special value):
- #for each set of variables, check if any variable contains at least one special value
- # m.16 <- sapply(special.v, function(x){
- # rowSums(sapply(c.16, function(y){
- # apply(dsa[, y, with=FALSE], 1, function(q){any(q==x)})
- # }))
- # })
-
- #(do this for each instance in c.16):
- #for each set of variables:
- # check if any variable contains at least one non-special value
- # + (for each special value) check if any variable contains at least special value
- m.16 <- lapply(c.16, function(x){
- cbind(apply(dsa[, x, with=FALSE], 1, function(q){any(!(q %in% special.v))}),
- sapply(special.v, function(y){
- apply(dsa[, x, with=FALSE], 1, function(q){any(q==y)})
- })
- )
- })
-
- # (do this for each instance in c.16)
- # if multiple special values per respondent exist, keep only the first one
- m.16 <- lapply(m.16, function(x){
- if(any(rowSums(x)>1)){
- p <- x[rowSums(x)>1,]
- for(i in 1:nrow(p)){
- a <- p[i,]
- f <- TRUE
- for(j in 1:length(a)){
- print(j)
- if(a[j] & f){
- f <- FALSE
- }else if(a[j] & !f){
- a[j] <- FALSE
- }
- }
- p[i,] <- a
- }
- x[rowSums(x)>1,] <- p
- }else{x}
- })
-
- m.n <- m.n + Reduce('+', m.16)
- }
-
- ##procedure for tip:17
- #only run if there is an at least one tip:17 variable
- if(survey.str[, any(tip==17)]){
- #get list of all unique tip:17 question ids
- q.17 <- unique(survey.str[tip==17, question.id])
-
- #get list of all corresponding variables for each q.17 id
- c.17 <- lapply(q.17, function(x){colnames(dsa)[which(colnames(dsa) %in% survey.str[question.id==x & tip==17, variable])]})
-
- #similiar procedure as for tip:2 and tip:16....
- m.17 <- lapply(c.17, function(x){
- cbind(apply(dsa[, x, with=FALSE], 1, function(q){any(!(q %in% special.v))}),
- sapply(special.v, function(y){
- apply(dsa[, x, with=FALSE], 1, function(q){any(q==y)})
- })
- )
- })
-
- #... the only difference is that we are checking for all rowsums > 0, not > 1
- m.17 <- lapply(m.17, function(x){
- if(any(rowSums(x)>1)){
- p <- x[rowSums(x)>0,]
- for(i in 1:nrow(p)){
- a <- p[i,]
- f <- TRUE
- for(j in 1:length(a)){
- if(a[j] & f){
- f <- FALSE
- }else if(a[j] & !f){
- a[j] <- FALSE
- }
- }
- p[i,] <- a
- }
- x[rowSums(x)>0,] <- p
- }else{x}
- })
-
- m.n <- m.n + Reduce('+', m.17)
- }
-
- m.n <- cbind(m.n, rowSums(m.n))
-
- if(all(m.n[, ncol(m.n)][1]==m.n[, ncol(m.n)])){
- m.n <- as.data.table(m.n)
- m.n[, recnum:=dsa$recnum]
- setnames(m.n, colnames(m.n)[-length(colnames(m.n))], c("va", "v1", "v3", "v5", "v96", "v97", "v98", "v99", "v4", "v2", "allqs"))
- setcolorder(m.n, c("recnum", colnames(m.n)[-length(colnames(m.n))]))
- return(m.n)
- }else{
- print("not all rowsums equal!")
- }
-}
\ No newline at end of file diff --git a/admin/survey/modules/mod_uporabnost/results/.gitignore b/admin/survey/modules/mod_uporabnost/results/.gitignore deleted file mode 100644 index a3a0c8b..0000000 --- a/admin/survey/modules/mod_uporabnost/results/.gitignore +++ /dev/null @@ -1,2 +0,0 @@ -*
-!.gitignore
\ No newline at end of file diff --git a/admin/survey/modules/mod_uporabnost/temp/.gitignore b/admin/survey/modules/mod_uporabnost/temp/.gitignore deleted file mode 100644 index a3a0c8b..0000000 --- a/admin/survey/modules/mod_uporabnost/temp/.gitignore +++ /dev/null @@ -1,2 +0,0 @@ -*
-!.gitignore
\ No newline at end of file |