Weird : cumsum not working on dplyr

Question

Context: I want to add cumulative sum column to my tibble named words_uni. I used library(dplyr), function mutate. I work with R version 3.4.1 64 bit - Windows 10 and RStudio Version 1.0.143

> head(words_uni)
# A tibble: 6 x 3
# Groups:   Type [6]
Type   Freq         per
<chr>  <int>       <dbl>
1   the 937839 0.010725848
2     i 918552 0.010505267
3    to 788892 0.009022376
4     a 615082 0.007034551

Then I did the following:

> words_uni1 = words_uni %>%
                      mutate( acum= cumsum(per))
> head(words_uni1)
# A tibble: 6 x 4
# Groups:   Type [6]
Type   Freq         per        acum
<chr>  <int>       <dbl>       <dbl>
1   the 937839 0.010725848 0.010725848
2     i 918552 0.010505267 0.010505267
3    to 788892 0.009022376 0.009022376
4     a 615082 0.007034551 0.007034551

Problem: It is not doing what I was expecting, and I cannot see why.

I would appreciate your comments. Thanks in advance.

Answer 1

You must have previously grouped the tibble by type. This causes your mutate call to calculate it by type.

Here is some reproducible code:

require(readr)
require(dplyr)

x <- read_csv("type, freq, per
the, 937839, 0.010725848
i, 918552, 0.010505267
to, 788892, 0.009022376
a, 615082, 0.007034551")


### ungrouped tibble, desired results
x %>% mutate(acum = cumsum(per))

# A tibble: 4 x 4
type   freq         per       acum
<chr>  <int>       <dbl>      <dbl>
1   the 937839 0.010725848 0.01072585
2     i 918552 0.010505267 0.02123112
3    to 788892 0.009022376 0.03025349
4     a 615082 0.007034551 0.03728804

### grouped tibble
x %>% group_by(type) %>% mutate(acum = cumsum(per))

# A tibble: 4 x 4
# Groups:   type [4]
type   freq         per        acum
<chr>  <int>       <dbl>       <dbl>
1   the 937839 0.010725848 0.010725848
2     i 918552 0.010505267 0.010505267
3    to 788892 0.009022376 0.009022376
4     a 615082 0.007034551 0.007034551

You need to simply ungroup your data.

word_uni %>% ungroup() %>% mutate(acum = cumsum(per))

Should do the trick.