Transpose / reshape dataframe without “timevar” fr

Question

I have a data frame that follows the below long Pattern:

   Name          MedName
  Name1    atenolol 25mg
  Name1     aspirin 81mg
  Name1 sildenafil 100mg
  Name2    atenolol 50mg
  Name2   enalapril 20mg

And would like to get below (I do not care if I can get the columns to be named this way, just want the data in this format):

   Name   medication1    medication2      medication3
  Name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
  Name2 atenolol 50mg enalapril 20mg             NA

Through this very site I have become familiarish with the reshape/reshape2 package, and have went through several attempts to try to get this to work but have thus far failed.

When I try dcast(dataframe, Name ~ MedName, value.var='MedName') I just get a bunch of columns that are flags of the medication names (values that get transposed are 1 or 0) example:

 Name  atenolol 25mg  aspirin 81mg
Name1              1             1
Name2              0             0 

I also tried a dcast(dataset, Name ~ variable) after I melted the dataset, however this just spits out the following (just counts how many meds each person has):

 Name  MedName
Name1        3
name2        2

Finally, I tried to melt the data and then reshape using idvar="Name" timevar="variable" (of which all just are Mednames), however this does not seem built for my issue since if there are multiple matches to the idvar, the reshape just takes the first MedName and ignores the rest.

Does anyone know how to do this using reshape or another R function? I realize that there probably is a way to do this in a more messy manner with some for loops and conditionals to basically split and re-paste the data, but I was hoping there was a more simple solution. Thank you so much!

Answer 1

@thelatemail's solution is similar to this one. When I generate the time variable, I use rle in case I'm not working interactively and the Name variable needs to be dynamic.

# start with your example data
x <- 
    data.frame(
        Name=c(rep("name1",3),rep("name2",2)),
        MedName=c("atenolol 25mg","aspirin 81mg","sildenafil 100mg",
            "atenolol 50mg","enalapril 20mg")
    )

# pick the id variable
id <- 'Name'

# sort the data.frame by that variable
x <- x[ order( x[ , id ] ) , ]

# construct a `time` variable on the fly
x$time <- unlist( lapply( rle( as.character( x[ , id ] ) )$lengths , seq_len ) )

# `reshape` uses that new `time` column by default
y <- reshape( x , idvar = id , direction = 'wide' )

# done
y

Answer 2

Here's a shorter way, taking advantage of the way unlist deals with names:

library(dplyr)
df1 %>% group_by(Name) %>% do(as_tibble(t(unlist(.[2]))))
# # A tibble: 2 x 4
# # Groups:   Name [2]
#      Name      MedName1       MedName2         MedName3
#     <chr>         <chr>          <chr>            <chr>
#   1 name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
#   2 name2 atenolol 50mg enalapril 20mg             <NA>

Answer 3

With the data.table package, this could easily be solved with the new rowid function:

library(data.table)
dcast(setDT(d1), 
      Name ~ rowid(Name, prefix = "medication"), 
      value.var = "MedName")

which gives:

   Name    medication1     medication2       medication3
1 Name1  atenolol 25mg    aspirin 81mg  sildenafil 100mg
2 Name2  atenolol 50mg  enalapril 20mg              <NA>

Another method (commonly used before version 1.9.7):

dcast(setDT(d1)[, rn := 1:.N, by = Name], 
      Name ~ paste0("medication",rn), 
      value.var = "MedName")

giving the same result.

---

A similar approach, but now using the dplyr and tidyr packages:

library(dplyr)
library(tidyr)
d1 %>%
  group_by(Name) %>%
  mutate(rn = paste0("medication",row_number())) %>%
  spread(rn, MedName)

which gives:

Source: local data frame [2 x 4]
Groups: Name [2]

    Name   medication1    medication2      medication3
  (fctr)         (chr)          (chr)            (chr)
1  Name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
2  Name2 atenolol 50mg enalapril 20mg               NA

Answer 4

You could always generate a unique timevar before using reshape. Here I use ave to apply the function seq_along 'along' each "Name".

test <- data.frame(
Name=c(rep("name1",3),rep("name2",2)),
MedName=c("atenolol 25mg","aspirin 81mg","sildenafil 100mg",
          "atenolol 50mg","enalapril 20mg")
)

# generate the 'timevar'
test$uniqid <- with(test, ave(as.character(Name), Name, FUN = seq_along))

# reshape!
reshape(test, idvar = "Name", timevar = "uniqid", direction = "wide")

Result:

   Name     MedName.1      MedName.2        MedName.3
1 name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
4 name2 atenolol 50mg enalapril 20mg             <NA>

Answer 5

This seems to actually be a fairly common problem, so I have included a function called getanID in my "splitstackshape" package.

Here's what it does:

library(splitstackshape)
getanID(test, "Name")
#     Name          MedName .id
# 1: name1    atenolol 25mg   1
# 2: name1     aspirin 81mg   2
# 3: name1 sildenafil 100mg   3
# 4: name2    atenolol 50mg   1
# 5: name2   enalapril 20mg   2

Since "data.table" is loaded along with "splitstackshape", you have access to dcast.data.table, so you can proceed as with @mnel's example.

dcast.data.table(getanID(test, "Name"), Name ~ .id, value.var = "MedName")
#     Name             1              2                3
# 1: name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
# 2: name2 atenolol 50mg enalapril 20mg               NA

---

The function essentially implements a sequence(.N) by the groups identified to create the "time" column.

Answer 6

Assuming your data is in the object dataset:

library(plyr)
## Add a medication index
data_with_index <- ddply(dataset, .(Name), mutate, 
                         index = paste0('medication', 1:length(Name)))    
dcast(data_with_index, Name ~ index, value.var = 'MedName')

##    Name   medication1    medication2      medication3
## 1 Name1 atenolol 25mg   aspirin 81mg sildenafil 100mg
## 2 Name2 atenolol 50mg enalapril 20mg             <NA>