CODEWITHSUNDEEP
rdplyr
Ahdee
Ahdee

Reputation: 4949

summarise_all by either mean or max depending on the value with dplyr?

Hi I have a dataframe 

d<- data.frame (type=c("rna","rna","rna"), value = c(1,2,3) )
d2 <- data.frame (type=c("dna","dna"), value = c(20,30) )
d3 <- data.frame (type=c("protein","protein","protein"), value = c(-9.6,300, 1000) )
df <- rbind (d,d2,d3)

     type  value
1     rna    1.0
2     rna    2.0
3     rna    3.0
4     dna   20.0
5     dna   30.0
6 protein   -9.6
7 protein  300.0
8 protein 1000.0

What I would like to do is to either use mean or max conditionally. Use max if there is even one value that that is < 0 else use mean. For example in this example the final df should look like this. 

 value    type
1   1000 protein
2    25     dna
3     2     rna

I tried to summarise as such but it errors out. 

df %>%
group_by(type) %>%
summarise_all(
    funs(
 if (. < 0 ){max}
 else{mean}
) )

Upvotes: 1

Views: 523

Answers (2)

akrun
akrun

Reputation: 887213

We can wrap it with any as the . < 0 is a logical vector of length greater than 1 and if/else works on a single TRUE/FALSE element. So, wrap with any to return that single element

df %>% 
   group_by(type) %>% 
   summarise_all(funs(if(any(. < 0)) max(.) else mean(.)))
# A tibble: 3 x 2
#  type    value
#  <fct>   <dbl>
#1 rna         2
#2 dna        25
#3 protein  1000

If we need to get the mean of only positive numbers

df %>% 
  group_by(type) %>% 
  summarise_all(funs(mean(.[.>= 0], na.rm = TRUE)))

NOTE: Here, we assume that in the original dataset, there are more numeric columns to get the mean of. It is better to add na.rm = TRUE where ever the parameter is there. If there are NA values in the dataset, it will remove it.

Upvotes: 2

jasbner
jasbner

Reputation: 2283

I think a regular summarise statement is more intuative in this situation.

df %>% 
  group_by(type) %>% 
  summarise(value = ifelse(any(value<0),max(value),mean(value)))

#  type      value
#  <fct>     <dbl>
#1 rna        2.00
#2 dna       25.0 
#3 protein 1000

Upvotes: 1

Related Questions