Why rbind throws a warning

Question

This is related to Are there more elegant ways to transform ragged data into a tidy dataframe

Why following code is not working:

events = structure(list(date = structure(c(-714974, -714579, -717835), class = "Date"), 
    days = c(1, 6, 0.5), name = c("Intro to stats", "Stats Winter school", 
    "TidyR tools"), topics = c("probability|R", "R|regression|ggplot", 
    "tidyR|dplyr")), .Names = c("date", "days", "name", "topics"
), row.names = c(NA, -3L), class = "data.frame")

> newdf <- data.frame(topic=character(), days=character())
> for(i in 1:length(events$topics)){
+ xx = unlist(strsplit(events$topics[i],'\\|'))
+ for(j in 1:length(xx)){
+ yy = c(xx[j], events$days[i]/length(xx))
+ print(yy)
+ newdf=rbind(newdf, yy)
+ }
+ }
[1] "probability" "0.5"        
[1] "R"   "0.5"
[1] "R" "2"
[1] "regression" "2"         
[1] "ggplot" "2"     
[1] "tidyR" "0.25" 
[1] "dplyr" "0.25" 
There were 11 warnings (use warnings() to see them)
> newdf
  X.probability. X.0.5.
1    probability    0.5
2           <NA>    0.5
3           <NA>   <NA>
4           <NA>   <NA>
5           <NA>   <NA>
6           <NA>   <NA>
7           <NA>   <NA>
> 
> warnings()
Warning messages:
1: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA ... :
  invalid factor level, NAs generated
2: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA,  ... :
  invalid factor level, NAs generated
3: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, 1L,  ... :
  invalid factor level, NAs generated
4: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA,  ... :
  invalid factor level, NAs generated
5: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, 1L,  ... :
  invalid factor level, NAs generated
6: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA,  ... :
  invalid factor level, NAs generated
7: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, 1L,  ... :
  invalid factor level, NAs generated
8: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA,  ... :
  invalid factor level, NAs generated
9: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, 1L,  ... :
  invalid factor level, NAs generated
10: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, NA,  ... :
  invalid factor level, NAs generated
11: In `[<-.factor`(`*tmp*`, ri, value = structure(c(1L, 1L,  ... :
  invalid factor level, NAs generated
> 

yy is okay but rbind is not working. Where is the error and how can it be corrected? Thanks for your help.

Answer 1

You may try:

newdf <- data.frame(topic=character(), daysPerTopic=character(), stringsAsFactors=F)
for(i in 1:length(events$topics)){
xx = unlist(strsplit(events$topics[i],'\\|'))
for(j in 1:length(xx)){
yy = data.frame(topic=xx[j], daysPerTopic=events$days[i]/length(xx), stringsAsFactors=F)
newdf <- rbind(newdf, yy) 
 }
 }

 newdf
#        topic daysPerTopic
# 1 probability         0.50
# 2           R         0.50
# 3           R         2.00
# 4  regression         2.00
# 5      ggplot         2.00
# 6       tidyR         0.25
# 7       dplyr         0.25

Or

 op <- options(stringsAsFactors=F)  #set to F

 #Your code
 newdf <- data.frame(topic=character(), days=character())
 for(i in 1:length(events$topics)){
 xx = unlist(strsplit(events$topics[i],'\\|'))
 for(j in 1:length(xx)){
yy = c(xx[j], events$days[i]/length(xx))
print(yy)
newdf=rbind(newdf, yy)
 }
 }

 newdf
#  X.probability. X.0.5.
# 1    probability    0.5
# 2              R    0.5
# 3              R      2
# 4     regression      2
# 5         ggplot      2
# 6          tidyR   0.25
# 7          dplyr   0.25

 options(op) #et back to default

Answer 2

Did you even try to debug your for loop? For example, by adding print(class(yy)) print(str(newdf)) you would see that after first iteration both newdf vectors become factors.

# [1] "probability" "0.5"        
# [1] "character"
# 'data.frame':  0 obs. of  2 variables:
#   $ topic: Factor w/ 0 levels: 
#   $ days : Factor w/ 0 levels: 
#   NULL
# [1] "R"   "0.5"
# [1] "character"
# 'data.frame': 1 obs. of  2 variables:
#   $ X.probability.: Factor w/ 1 level "probability": 1
# $ X.0.5.        : Factor w/ 1 level "0.5": 1
# NULL
# [1] "R" "2"
# [1] "character"
# 'data.frame': 2 obs. of  2 variables:
#   $ X.probability.: Factor w/ 1 level "probability": 1 NA
# $ X.0.5.        : Factor w/ 1 level "0.5": 1 1

...

You would say "but I defined them as character". True, but if you'll read rbind documentation, you will see that

For cbind (rbind), vectors of zero length (including NULL) are ignored unless the result would have zero rows (columns), for S compatibility. (Zero-extent matrices do not occur in S3 and are not ignored in R.)

Another property of rbind is that it inherits it's properties from data.frame while one of them is stringsAsFactors == TRUE

What happened here could be easily illustrated in a dummy example, consider

temp <- data.frame(A = letters[1:3])
str(temp)
## 'data.frame':    3 obs. of  1 variable:
## $ A: Factor w/ 3 levels "a","b","c": 1 2 3

temp$A[3] <- "d"
## Warning message:
## In `[<-.factor`(`*tmp*`, 3, value = c(1L, 2L, NA)) :
##   invalid factor level, NA generated

temp$A
## [1] a    b    <NA>
## Levels: a b c

You can see two things here:

• data.frame automatically converted character class to factors
• When trying to parse a new level to factor vector it converts it into NA and throws the exact error you were receiving

As mentioned by @akrun, setting to options(stringsAsFactors=F) will solve your problem

Answer 3

Set options(stringsAsFactors=FALSE) and your code should work as expected. The reason for the warnings and NA's in the result is because of the implicit conversion to factors and the type mismatch between newdf columns and yy, see https://stackoverflow.com/a/1640729/1541036.

For a cleaner way of achieving the same result, here's a group by solution using data.table

library(data.table)
events <- as.data.table(events)
events2 <- events[, list(topic=unlist(strsplit(topics, '|', fixed=TRUE))), by=c("date", "days", "name")]
events2[, probability := days / .N, by=name]