##setwd("/media/49ED-21FF/REVERSE/wd")
tbtest<-read.csv("pb1.csv",row.names=1)

str(tbtest)
tbtest$gender<-as.factor(tbtest$gender)
tbtest$atcd1<-as.factor(tbtest$atcd1)
tbtest$atcd2<-as.factor(tbtest$atcd2)
tbtest$atcd3<-as.factor(tbtest$atcd3)
tbtest$atcd4<-as.factor(tbtest$atcd4)
tbtest$f1<-as.factor(tbtest$f1)
tbtest$f2<-as.factor(tbtest$f2)
tbtest$f3<-as.factor(tbtest$f3)
tbtest$f4<-as.factor(tbtest$f4)
tbtest$f6<-as.factor(tbtest$f6)
tbtest$f7<-as.factor(tbtest$f7)
tbtest$f8<-as.factor(tbtest$f8)
tbtest$f678<-as.factor(tbtest$f678)
tbtest$f9<-as.factor(tbtest$f9)
tbtest$f10<-as.factor(tbtest$f10)
tbtest$f10a<-as.factor(tbtest$f10a)
tbtest$f11<-as.factor(tbtest$f11)
str(tbtest)

## where :
# f1a<-"H" if (gender=="0")
#    <-"Favecco" if(gender=="1 & f1=="TRUE")
#    <-"Fsansco" if(gender=="1 & f1=="FALSE")
# with(tbtest,ftable(f1a~gender+f1))
# f2a<-"H" if (gender=="0")
#    <-"Favechrt" if(gender=="1 & f2=="TRUE")
#    <-"Fsanshrt" if(gender=="1 & f2=="FALSE")
# with(tbtest,ftable(f2a~gender+f2)
# f12a<-"H" if (gender=="0")
#     <-"Favechorm" if(gender=="1 & (f1=="TRUE" | f2=="TRUE"))
#     <-"Fsanshorm" if(gender=="1 & (f1=="FALSE" & f2=="FALSE"))
# f678<-"1" if (f6=="1"|f7=="1"|f8=="1")

# c9a is c9 with some NA values set to 1 (theoritical range 0-1 : 0-100%) (not yet used : particular distribution)
# f10a is f10 with a new level : some NA values set to "0" (not yet used because of a small group : should relevel)

library(Hmisc)

## to generate "Error in matxv(X, xcof) : columns in a (46) must be <= length of b (44)"
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
  +f1a+f2a+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8+f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## with "I()" for continuous variables : same error, different numbers
## "Error in matxv(X, xcof) : columns in a (33) must be <= length of b (31)"
tbtestI<-aregImpute(~I(time)+event
  +gender+I(age)+I(weight)+I(height)+I(bmi)
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
  +f1a+f2a+f12a
  +f3+f4
  +I(c1)
  +I(c2)+I(c3)+I(c4)
  +I(c5)+I(c6)+I(c7)+I(c8)
  +f6+f7+f8+f678
  +f9
#  +I(c9)+f10
#  +I(c9a)+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## commenting f1a+f2a+f12a solves the error
## but unmasks warnings : "In f$xcoef[, 1] * f$xcenter..."
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
#  +f1a+f2a+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8+f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## uncommenting f1+f2 is ok (with a fifth warning)
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
  +f1+f2
#  +f1a+f2a+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8+f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## ignoring gender (f1a f2a definitions) doesn't change anything 
tbtestI<-aregImpute(~time+event
#  +gender
  +age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
  +f1a+f2a+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8+f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## ignoring f12a doesn't change anything 
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
  +f1a+f2a #+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8+f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## ignoring f1a and f2a (and not f12a) doesn't change anything 
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
#  +f1+f2
#  +f1a+f2a
  +f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8 #f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

## ignoring f678 reduces number of warnings
tbtestI<-aregImpute(~time+event
  +gender+age+weight+height+bmi
  +atcd1+atcd2
  +atcd3+atcd4
  +f1+f2
#  +f1a+f2a+f12a
  +f3+f4
  +c1
  +c2+c3+c4
  +c5+c6+c7+c8
  +f6+f7+f8 #f678
  +f9
#  +c9+f10
#  +c9a+f10a
  +f11,
  n.impute=5,data=tbtest
  )

#So I guess f1a f2a f12a are the problem but I don't understand why
#(c9/c9a is another problem : a percentage I should probably transform before imputing)