通过模式匹配熔化列

Question

我有一个非常广泛的数据框，其中包含标准的人口统计特征（年龄、性别、种族、教育、收入等）。我还有受访者对问题的回答，这些问题可以用四种方式之一（使用 "cb"、"lb"、"lw" 或 "cw"）作为序言。

数据框目前采用宽格式，每一行代表单个受访者的答案。我想将其转换为长格式，但我找不到使用 reshape2 库的直接解决方案。

我想将所有人口统计特征保留为它们自己的列，但将问题、答案、置信度和分数列减少到它们自己的融合列中。这是我正在寻找的东西的想法：

string <- "
response_id,age,sex,race_1,race_2,race_3,cb_1,cb_1_conf,cb_1_ans,cb_1_score,lb_1,lb_1_conf,lb_1_ans,lb_1_score
11,25,M,white,NA,NA,Astrophysicist,9,Dog,0,Jackson,8,Jackson,1
22,27,F,NA,black,asian,Monkey,8,Dog,0,Jackson,7,Jackson,1"

x <- read.csv(con <- textConnection(string), header=TRUE)

看起来像这样：

> x
  response_id age sex race_1 race_2 race_3           cb_1 cb_1_conf cb_1_ans cb_1_score    lb_1 lb_1_conf lb_1_ans lb_1_score
1          11  25   M  white   <NA>     NA Astrophysicist         9      Dog          0 Jackson         8  Jackson          1
2          22  27   F   <NA>  black     NA         Monkey         8      Dog          0 Jackson         7  Jackson          1

希望将其转换成这种形式：

string_2 <- "
response_id,age,sex,race,question,response,confidence,correct_answer,score
11,25,M,white,cb_1,Astrophysicist,9,Dog,0
11,25,M,white,lb_1,Jackson,8,Jackson,1
22,27,F,black/asian,cb_1,Monkey,8,Dog,0
22,27,F,black/asian,lb_1,Jackson,8,Jackson,1
"

x_2 <- read.csv(con <- textConnection(string_2), header=TRUE)

  response_id age sex        race question       response confidence correct_answer score
1          11  25   M       white     cb_1 Astrophysicist          9            Dog     0
2          11  25   M       white     lb_1        Jackson          8        Jackson     1
3          22  27   F black/asian     cb_1         Monkey          8            Dog     0
4          22  27   F black/asian     lb_1        Jackson          8        Jackson     1

我尝试对 df 进行子集化以仅包含以 cb、lb、cw 或 lw 为前缀的列，然后是：

    melt(subset, id=c("ResponseID"), 
    +      measure.vars=grep("^(CB|LB|LW|CW)", colnames(subset)))

But this doesn't allow me to flexibly melt the _conf columns the _ans columns and the _score columns.  

I had to modify Maurits' answer a bit to work better for my case.  Here is my solution:

df_test <- df_ans %>%
    unite(race, contains("race"), sep = "/") %>%                # combine race_1,2,3
    mutate(race = str_replace_all(race, "(/NA|NA/)", "")) %>%   # replace NA from race
    select_all( ~ gsub("(^[A-Z][A-Z]_\d+$)", "\1_response", .)) %>%  # add "_response" to Q
    gather(key, val, -(1:24)) %>%                                # wide to long
    separate(key, c("q1", "q2", "item")) %>%                    # split into Q + item
    unite(question, q1, q2, sep = "_") %>%                      # [continued]
    mutate(item = gsub("_", "", item)) %>%                      # [continued]
    spread(item, val) %>%                                       # long to wide
    rename(answer = ans, confidence = con)                      # rename columns

Answer 1

这是一个tidyverse解决方案：

x %>%
    unite(race, contains("race"), sep = "/") %>%                # combine race_1,2,3
    mutate(race = str_replace_all(race, "(/NA|NA/)", "")) %>%   # replace NA from race
    select_all( ~ gsub("^(\w+_\d)$", "\1_response", .)) %>%  # add "_response" to Q
    gather(key, val, -(1:4)) %>%                                # wide to long
    separate(key, c("q1", "q2", "item")) %>%                    # split into Q + item
    unite(question, q1, q2, sep = "_") %>%                      # [continued]
    mutate(item = gsub("_", "", item)) %>%                      # [continued]
    spread(item, val) %>%                                       # long to wide
    rename(answer = ans, confidence = conf)                     # rename columns
#  response_id age sex        race question  answer confidence       response
#1          11  25   M       white     cb_1     Dog          9 Astrophysicist
#2          11  25   M       white     lb_1 Jackson          8        Jackson
#3          22  27   F black/asian     cb_1     Dog          8         Monkey
#4          22  27   F black/asian     lb_1 Jackson          7        Jackson
#  score
#1     0
#2     1
#3     0
#4     1

解释：

根据 race_1,race_2,race_3, whilst removingNA`s 中的条目创建 united race .
剩下的就是 gathering、spreading 和 separateing 条目以分离出 question、answer、confidence 和 response.
我在这里假设所有问题的形式都是\w+_\d（例如cb_1、lb_1）；必要时进行调整。

通过模式匹配熔化列

Melting columns by pattern matching

r

melt

dplyr

tidyr

tidyverse