melt + strsplit, or opposite to aggregate

Question

I have a little question that seems to be so easy in concept, but I cannot find the way to do it...

Say I have a data.frame df2 with a column listing car brands and another column with all the models per brand separated by ','. I have obtained df2 aggregating another data.frame named df1 with the primary key being the model.

How should I proceed to do the opposite task (i.e.: from df2 to df1)? My guess is something like melt(df2, id=unlist(strsplit('models',',')))... Many thanks!

Here is a MWE:

df1 <- data.frame(model=c('a1','a2','a3','b1','b2','c1','d1','d2','d3','d4'), 
                      brand=c('a','a','a','b','b','c','d','d','d','d'))
df1
collap <- function(x){
  out <- paste(sort(unique(x)), collapse=",")
  return (out)
}
df2 <- aggregate(df1$model, by=list(df1$brand), collap)
names(df2) <- c('brand','models')
df2 #how can I do the opposite task (ie: from df2 to df1)?

Answer 1

These days I would use tidytext::unnest_tokens for this task:

library(tidytext)
df2 %>% 
  unnest_tokens(model, models, token = "regex", pattern = ",")

# A tibble: 10 x 2
    brand model
   <fctr> <chr>
 1      a    a1
 2      a    a2
 3      a    a3
 4      b    b1
 5      b    b2
 6      c    c1
 7      d    d1
 8      d    d2
 9      d    d3
10      d    d4

Answer 2

Here are two alternatives:

Use data.table and unlist as follows:

library(data.table)
DT <- data.table(df2)
DT[, list(model = unlist(strsplit(as.character(models), ","))), 
   by = brand]
#     brand model
#  1:     a    a1
#  2:     a    a2
#  3:     a    a3
#  4:     b    b1
#  5:     b    b2
#  6:     c    c1
#  7:     d    d1
#  8:     d    d2
#  9:     d    d3
# 10:     d    d4

Use concat.split.multiple from my "splitstackshape" package. One nice thing with this approach is being able to split multiple columns with one simple command.

library(splitstackshape)
out <- concat.split.multiple(df2, "models", ",", "long")
out[complete.cases(out), ]
#    brand time models
# 1      a    1     a1
# 2      b    1     b1
# 3      c    1     c1
# 4      d    1     d1
# 5      a    2     a2
# 6      b    2     b2
# 8      d    2     d2
# 9      a    3     a3
# 12     d    3     d3
# 16     d    4     d4

Answer 3

Playing around, I have found a way to do the trick, even though it may be quite dirty:

df1 <- data.frame(model=as.character(melt(strsplit(df2$models,','))$value), brand=as.character(df2[match(melt(strsplit(df2$models,','))$L1, rownames(df2)),]$brand))

It is not the best solution, since the data.frames actually have many more columns, and I would not want to go one by one... If someone knows a prettier way to solve this, I would appreciate it!

Answer 4

Here is how I would do it using the plyr package

library("plyr")
ddply(df2, .(brand), function(DF) {
  data.frame(model = strsplit(DF$models, ",")[[1]])
})

As a point of comparison, this is how to use the same package to go from df1 to df2:

ddply(df1, .(brand), 
      summarize, models=paste(sort(unique(model)), collapse=","))