R：列集中的最后一个非NA值

Question

I am looking for a solution to the problem below that would be supported in pipes.

I have data that looks like this:

tibble(
  column_set_1_1 = c(1, 2, 3), column_set_1_2 = c(2, 3, NA), column_set_1_3 = c(3, NA, NA),
  column_set_2_1 = c(1, 2, 3), column_set_2_2 = c(4, 5, 6), column_set_2_3 = c(7, 8, 9), 
  column_set_2_4 = c(10, 11, NA), column_set_2_5 = c(13, NA, NA), column_set_2_6 = c(NA, NA, NA)
)

# A tibble: 3 × 9
  column_set_1_1 column_set_1_2 column_set_1_3 column_set_2_1 column_set_2_2 column_set_2_3 column_set_2_4 column_set_2_5 column_set_2_6
           <dbl>          <dbl>          <dbl>          <dbl>          <dbl>          <dbl>          <dbl>          <dbl>          <lgl>
1              1              2              3              1              4              7             10             13             NA
2              2              3             NA              2              5              8             11             NA             NA
3              3             NA             NA              3              6              9             NA             NA             NA

I am basically looking to get the last non-NA value by column set. So, the expected output is:

tibble(
  column_set_1 = c(3, 3, 3), 
  column_set_2 = c(13, 11, 9)
)

# A tibble: 3 × 2
  column_set_1 column_set_2
         <dbl>        <dbl>
1            3           13
2            3           11
3            3            9

Answer 1

Here is a tidyverse approach without reshaping the original data frame but split it into groups by the column names pattern, and use coalesce function to get the last non-NA values in each sub data frame:

library(tidyverse)
df_foo %>% 
      mutate_all(as.numeric) %>% 
      split.default(f = sub("_\\d+$", "", names(.))) %>% 
      map_df(~do.call(coalesce, setNames(rev(.), NULL)))

# A tibble: 3 × 2
#  column_set_1 column_set_2
#         <dbl>        <dbl>
#1            3           13
#2            3           11
#3            3            9

Answer 2

Here is my solution using tidyverse tools:

library(dplyr)
library(tidyr)
library(stringr)
library(tibble)

get_last_nonNA <- function(vec) {
  return(last(vec[!is.na(vec)]))
}

convert_table_last_nonNA <- . %>%
  rownames_to_column() %>%
  gather(key=column_type, value=value, -rowname) %>%
  mutate(column_set=str_extract(string=column_type,
                                pattern="[0-9]+")) %>%
  group_by(column_set, rowname) %>%
  summarise(last_nonNA_value=get_last_nonNA(value)) %>%
  spread(key=column_set, value=last_nonNA_value) %>%
  select(-rowname) %>%
  select(colnames(.) %>% as.integer() %>% order()) %>%
  "colnames<-"(paste0("column_set_", colnames(.)))
# Usage
data_tbl <- tibble(
  column_set_1_1 = c(1, 2, 3), column_set_1_2 = c(2, 3, NA),
  column_set_1_3 = c(3, NA, NA), column_set_2_1 = c(1, 2, 3),
  column_set_2_2 = c(4, 5, 6), column_set_2_3 = c(7, 8, 9), 
  column_set_2_4 = c(10, 11, NA), column_set_2_5 = c(13, NA, NA),
  column_set_2_6 = c(NA, NA, NA)
)

convert_table_last_nonNA(data_tbl)

# # A tibble: 3 × 2
#   column_set_1 column_set_2
# *        <dbl>        <dbl>
# 1            3           13
# 2            3           11
# 3            3            9

What it does, step by step:

1. Creates a reusable pipe with convert_table_last_nonNA <- . %>% convert_table_last_nonNA <- . %>% ;
2. Adds row names to the separate column with rownames_to_column() in order to have information for extracting the last non-NA data per row;
3. Transforms input table into long format with gather(key=column_type, value=value, -rowname) : the rows represent now a combination of key columns ( rowname and column_type ) and value ( value );
4. Computes column's set number via regular expression magic (extracts the first number from column_type strings) and stores it in separate column column_set . This is done with mutate(column_set=str_extract(string=column_type, pattern="[0-9]+")) ;
5. Summarises the data in needed fashion with group_by(column_set, rowname) %>% summarise(last_nonNA_value=get_last_nonNA(value)) . That is "for every combination of column_set and rowname give the last nonNA value of value (via get_last_nonNA call) and stores it in column last_nonNA_value ". Note : if there are only NA 's for some combination of column_set and rowname the result will be NA;
6. Transforms table in wide format with spread(key=column_set, value=last_nonNA_value) . Now there is a column for every item in column_set and their values are last_nonNA_value s;
7. Deletes column rowname because it is not needed any more;
8. Reorders columns in order of increasing number of column_set. It is needed because if there are more then 9 column sets in your original data then there will be some confusion with ordering columns (that is column column_set_10 will be placed directly after column_set_1 ). This is done with select(colnames(.) %>% as.integer() %>% order()) ;
9. Adds prefix column_set_ to column names with "colnames<-"(paste0("column_set_", colnames(.))) .

Answer 3

Here is a solution that I came up with that works with pipes:

df_foo %>% 
  gather(key = Key, value = Value, -ID) %>% 
  mutate(set = str_extract(Key, "column_set_[0-9]")) %>% 
  mutate(number = str_extract(Key, "(?<=column_set_[0-9]_)[0-9]+")) %>% 
  group_by(ID, set) %>% 
  dplyr::filter(!is.na(Value)) %>%
  arrange(number) %>% 
  slice(n()) %>% 
  select(-number, -Key) %>% 
  spread(key = set, value = Value)

I don't like the fact that I have to arrange and then slice out the last row -- seems inelegant to me. Any improvements welcome.