8

Let's say I have a tibble.

library(tidyverse) 
tib <- as.tibble(list(record = c(1:10), 
                      gender = as.factor(sample(c("M", "F"), 10, replace = TRUE)), 
                      like_product = as.factor(sample(1:5, 10, replace = TRUE)))
tib

    # A tibble: 10 x 3
   record gender like_product
    <int> <fctr>       <fctr>
 1      1      F            2
 2      2      M            1
 3      3      M            2
 4      4      F            3
 5      5      F            4
 6      6      M            2
 7      7      F            4
 8      8      M            4
 9      9      F            4
10     10      M            5

I would like to dummy code my data with 1's and 0's so that the data looks more/less like this.

# A tibble: 10 x 8
   record gender_M gender_F like_product_1 like_product_2 like_product_3 like_product_4 like_product_5
    <int>    <dbl>    <dbl>          <dbl>          <dbl>          <dbl>          <dbl>          <dbl>
 1      1        0        1              0              0              1              0              0
 2      2        0        1              0              0              0              0              0
 3      3        0        1              0              1              0              0              0
 4      4        0        1              1              0              0              0              0
 5      5        1        0              0              0              0              0              0
 6      6        0        1              0              0              0              0              0
 7      7        0        1              0              0              0              0              0
 8      8        0        1              0              1              0              0              0
 9      9        1        0              0              0              0              0              0
10     10        1        0              0              0              0              0              1

My workflow would require that I know a range of variables to dummy code (i.e. gender:like_product), but don't want to identify EVERY variable by hand (there could be hundreds of variables). Likewise, I don't want to have to identify every level/unique value of every variable to dummy code. I'm ultimately looking for a tidyverse solution.

I know of several ways of doing this, but none of them that fit perfectly within tidyverse. I know I could use mutate...

tib %>%
     mutate(gender_M = ifelse(gender == "M", 1, 0), 
            gender_F = ifelse(gender == "F", 1, 0), 
            like_product_1 = ifelse(like_product == 1, 1, 0), 
            like_product_2 = ifelse(like_product == 2, 1, 0), 
            like_product_3 = ifelse(like_product == 3, 1, 0), 
            like_product_4 = ifelse(like_product == 4, 1, 0), 
            like_product_5 = ifelse(like_product == 5, 1, 0)) %>%
     select(-gender, -like_product)

But this would break my workflow rules of needing to specify every dummy coded output.

I've done this in the past with model.matrix, from the stats package.

model.matrix(~ gender + like_product, tib)

Easy and straightforward, but I want a solution in the tidyverse. EDIT: Reason being, I still have to specify every variable, and being able to use select helpers to specify something like gender:like_product would be much preferred.

I think the solution is in purrr

library(purrr)
dummy_code <- function(x) {
     lvls <- levels(x)
     sapply(lvls, function(y) as.integer(x == y)) %>% as.tibble
} 

tib %>%
     map_at(c("gender", "like_product"), dummy_code)

$record
 [1]  1  2  3  4  5  6  7  8  9 10

$gender
# A tibble: 10 x 2
       F     M
   <int> <int>
 1     1     0
 2     0     1
 3     0     1
 4     1     0
 5     1     0
 6     0     1
 7     1     0
 8     0     1
 9     1     0
10     0     1

$like_product
# A tibble: 10 x 5
     `1`   `2`   `3`   `4`   `5`
   <int> <int> <int> <int> <int>
 1     0     1     0     0     0
 2     1     0     0     0     0
 3     0     1     0     0     0
 4     0     0     1     0     0
 5     0     0     0     1     0
 6     0     1     0     0     0
 7     0     0     0     1     0
 8     0     0     0     1     0
 9     0     0     0     1     0
10     0     0     0     0     1

This attempt produces a list of tibbles, with the exception of the excluded variable record, and I've been unsuccessful at combining them all back into a single tibble. Additionally, I still have to specify every column, and overall it seems clunky.

Any better ideas? Thanks!!

user438383
  • 5,716
  • 8
  • 28
  • 43
Jacob Nelson
  • 443
  • 1
  • 6
  • 16
  • If using `model.matrix` is __"easy and straightforward"__ why should someone on this site spend time coming up with a `tidyverse` solution? `model.matrix` does exactly what you want. Why should _everything_ have / need a `tidyverse` solution? – bouncyball Mar 22 '18 at 16:58
  • 1
    "Easy and straightforward, but I want a solution in the tidyverse." Why? You could just do `%>% model.matrix(~ x + y, data = .) %>% as.data_frame()`. – Nathan Werth Mar 22 '18 at 16:58
  • Another useful one is `dummy.data.frame` from package `dummies`. See here: https://www.rdocumentation.org/packages/dummies/versions/1.5.6/topics/dummy – AntoniosK Mar 22 '18 at 17:01
  • 2
    Fair points...but notice that model.matrix still requires I specify every variable. I suppose I could parse a workaround, but it wouldn't be very pretty. I suspected that there was a way to do it in tidyverse, and it could be useful to the community...it's a common enough data transformation that there *ought to be* a tidyverse way to do it, that doesn't require writing out class type transformations with `%>% as.tibble()`, and can take advantage of intuitive syntax and tools like select-helpers. I still think this is a good question. – Jacob Nelson Mar 22 '18 at 17:43
  • `do.call('cbind', tib %>% map_at(c("gender", "like_product"), dummy_code))` or `tib %>% map_at(c("gender", "like_product"), dummy_code) %>% reduce(cbind.data.frame)` and play with the names? – bouncyball Mar 22 '18 at 18:43

2 Answers2

10

An alternative to model.matrix is using the package recipes. This is still a work in progress and is not yet included in the tidyverse. At some point it might / will be included in the tidyverse packages.

I will leave it up to you to read up on recipes, but in the step step_dummy you can use special selectors from the tidyselect package (installed with recipes) like the selectors you can use in dplyr as starts_with(). I created a little example to show the steps.

Example code below.

But if this is handier I will leave up to you as this has already been pointed out in the comments. The function bake() uses model.matrix to create the dummies. The difference is mostly in the column names and of course in the internal checks that are being done in the underlying code of all the separate steps.

library(recipes)
library(tibble)

tib <- as.tibble(list(record = c(1:10), 
                      gender = as.factor(sample(c("M", "F"), 10, replace = TRUE)), 
                      like_product = as.factor(sample(1:5, 10, replace = TRUE))))

dum <- tib %>% 
  recipe(~ .) %>% 
  step_dummy(gender, like_product) %>% 
  prep(training = tib) %>% 
  bake(newdata = tib)

dum

# A tibble: 10 x 6
   record gender_M like_product_X2 like_product_X3 like_product_X4 like_product_X5
    <int>    <dbl>           <dbl>           <dbl>           <dbl>           <dbl>
 1      1       1.              1.              0.              0.              0.
 2      2       1.              1.              0.              0.              0.
 3      3       1.              1.              0.              0.              0.
 4      4       0.              0.              1.              0.              0.
 5      5       0.              0.              0.              0.              0.
 6      6       0.              1.              0.              0.              0.
 7      7       0.              1.              0.              0.              0.
 8      8       0.              0.              0.              1.              0.
 9      9       0.              0.              0.              0.              1.
10     10       1.              0.              0.              0.              0.
phiver
  • 23,048
  • 14
  • 44
  • 56
  • 2
    This is EXACTLY what I'm looking for. This is intuitive, preserves good names, and uses tidy selectors. I'll have to read up more on the `recipes` package. Thanks! – Jacob Nelson Mar 22 '18 at 19:06
0

In case you don't want to load any additional packages, you could also use pivot_wider statements like this:

tib %>% 
  mutate(dummy = 1) %>% 
  pivot_wider(names_from = gender, values_from = dummy, values_fill = 0) %>%
  mutate(dummy = 1) %>% 
  pivot_wider(names_from = like_product, values_from = dummy, values_fill = 0, names_glue = "like_product_{like_product}")
JRSlotman
  • 1
  • 1