6 min read

naniar 使用技巧 缺失值展示

2018/11/03

本文于2020-10-10更新。 如发现问题或者有建议，欢迎提交 Issue

library(visdat)
library(tidyverse)
library(naniar)
library(rmarkdown)

1 缺失值位置

vis_dat(airquality) +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

vis_miss(airquality) +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

  1. 缺点是不能进行ggplot2函数的叠加。 ( Github )

2 缺失值分布

One approach to visualising missing data comes from ggobi and manet, where we replace “NA” values with values 10% lower than the minimum value in that variable. (Tierney 2018c)

缺失值的展示方法中，一种是使用最小值还要小10%的值来代替，这样就可以进行散点图等方式进行展示。

ggplot(airquality, 
       aes(x = Solar.R, 
           y = Ozone)) + 
    geom_miss_point() +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

3 缺失值数量

gg_miss_var(airquality) +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

4 样本缺失值情况

miss_case_summary(airquality)
## # A tibble: 153 x 3
##     case n_miss pct_miss
##    <int>  <int>    <dbl>
##  1     5      2     33.3
##  2    27      2     33.3
##  3     6      1     16.7
##  4    10      1     16.7
##  5    11      1     16.7
##  6    25      1     16.7
##  7    26      1     16.7
##  8    32      1     16.7
##  9    33      1     16.7
## 10    34      1     16.7
## # ... with 143 more rows
  1. case是行的index
  2. n_miss是某行缺失值的个数，pct_miss是某行缺失值的比例。
miss_case_table(airquality)
## # A tibble: 3 x 3
##   n_miss_in_case n_cases pct_cases
##            <int>   <int>     <dbl>
## 1              0     111     72.5 
## 2              1      40     26.1 
## 3              2       2      1.31

再按照n_miss_in_case进行了汇总。

4.1 样本缺失值预测

根据miss_case_summary可以知道每个样本的缺失情况，可以对这个缺失率或者值，进行模型预测，看哪个变量比较显著。

library(rpart)
library(rpart.plot)
## Warning: 程辑包'rpart.plot'是用R版本3.6.3 来建造的
airquality %>%
    add_prop_miss() %>%
    rpart(prop_miss_all ~ ., data = .) %>%
    prp(type = 4, extra = 101, prefix = "Prop. Miss = ")    
## Warning: Cannot retrieve the data used to build the model (so cannot determine roundint and is.binary for the variables).
## To silence this warning:
##     Call prp with roundint=FALSE,
##     or rebuild the rpart model with model=TRUE.

5 变量缺失值情况

miss_var_summary(airquality)
## # A tibble: 6 x 3
##   variable n_miss pct_miss
##   <chr>     <int>    <dbl>
## 1 Ozone        37    24.2 
## 2 Solar.R       7     4.58
## 3 Wind          0     0   
## 4 Temp          0     0   
## 5 Month         0     0   
## 6 Day           0     0
miss_var_table(airquality)
## # A tibble: 3 x 3
##   n_miss_in_var n_vars pct_vars
##           <int>  <int>    <dbl>
## 1             0      4     66.7
## 2             7      1     16.7
## 3            37      1     16.7
pedestrian %>%
    group_by(month) %>%
    miss_var_summary() %>%
    filter(variable == "hourly_counts")
## # A tibble: 12 x 4
## # Groups:   month [12]
##    month     variable      n_miss pct_miss
##    <ord>     <chr>          <int>    <dbl>
##  1 January   hourly_counts      0     0   
##  2 February  hourly_counts      0     0   
##  3 March     hourly_counts      0     0   
##  4 April     hourly_counts    552    19.2 
##  5 May       hourly_counts     72     2.42
##  6 June      hourly_counts      0     0   
##  7 July      hourly_counts      0     0   
##  8 August    hourly_counts    408    13.7 
##  9 September hourly_counts      0     0   
## 10 October   hourly_counts    412     7.44
## 11 November  hourly_counts    888    30.8 
## 12 December  hourly_counts    216     7.26

5.1 interaction 展示

主要为了查看变量之间产生缺失值的共轭现象。 (Tierney 2018b)

gg_miss_upset(riskfactors)

gg_miss_upset(riskfactors, 
              nsets = 10,
              nintersects = 50)

6 填补缺失值

library(simputation)
## Warning: 程辑包'simputation'是用R版本3.6.3 来建造的
## 
## 载入程辑包：'simputation'
## The following object is masked from 'package:naniar':
## 
##     impute_median
ocean_imp <- oceanbuoys %>%
    bind_shadow() %>%
    impute_lm(air_temp_c ~ wind_ew + wind_ns) %>%
    impute_lm(humidity ~  wind_ew + wind_ns) %>%
    impute_lm(sea_temp_c ~  wind_ew + wind_ns) %>%
    add_label_shadow() %>% 
    paged_table()

add_label_shadow函数打上标记any_missing。 (Tierney 2018a)

library(ggplot2)
ggplot(ocean_imp,
    aes(x = air_temp_c,
       y = humidity,
       color = any_missing)) + 
    geom_point() +
    scale_color_brewer(palette = "Dark2") +
    theme(legend.position = "bottom") +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

ggplot(ocean_imp,
       aes(x = air_temp_c,
           fill = any_missing)) + 
    geom_density(alpha = 0.3) + 
    scale_fill_brewer(palette = "Dark2") +
    theme(legend.position = "bottom") +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

ggplot(ocean_imp,
       aes(x = humidity,
           fill = any_missing)) + 
    geom_density(alpha = 0.3) + 
    scale_fill_brewer(palette = "Dark2") +
    theme(legend.position = "bottom") +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

7 Cases

library(data.table)
library(tidyverse)
library(visdat)
library(naniar)
# data <- fread('ldfilter_cbind.txt')
data <- fread(here::here('../tutoring2/pansiyu/analysis/NA_inlm/ldfilter_cbind.txt'))
dim(data)
## [1] 683  53

缺失值处理参考 naniar 使用技巧 缺失值展示 。

vis_miss(data) +
    theme(text = element_text(size=10),
    axis.text.x = element_text(angle=90, hjust=1)) +
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

联动缺失不高。

联动缺失不高，点击前面的三角查看更多。

gg_miss_upset(data)

gg_miss_upset(data, 
              nsets = 10,
              nintersects = 50)

样本缺失率不高。

miss_case_summary(data)
## # A tibble: 683 x 3
##     case n_miss pct_miss
##    <int>  <int>    <dbl>
##  1   106      5     9.43
##  2   154      4     7.55
##  3    10      3     5.66
##  4    15      3     5.66
##  5    26      3     5.66
##  6    74      3     5.66
##  7   129      3     5.66
##  8   213      3     5.66
##  9   231      3     5.66
## 10   372      3     5.66
## # ... with 673 more rows
miss_case_table(airquality)
## # A tibble: 3 x 3
##   n_miss_in_case n_cases pct_cases
##            <int>   <int>     <dbl>
## 1              0     111     72.5 
## 2              1      40     26.1 
## 3              2       2      1.31
  1. 最多一个样本，缺失值也就5个。

变量缺失率不高。

gg_miss_var(data) +
    theme(
        text = element_text(size=8))+
    labs(
        caption = "Jiaxiang Li - jiaxiangli.netlify.com"
    )

miss_var_summary(data)
## # A tibble: 53 x 3
##    variable      n_miss pct_miss
##    <chr>          <int>    <dbl>
##  1 rs2349573_G       21     3.07
##  2 rs143104263_A     21     3.07
##  3 rs201145_G        20     2.93
##  4 rs804553_C        20     2.93
##  5 rs370245477_C     18     2.64
##  6 rs201569_T        18     2.64
##  7 rs141184610_A     17     2.49
##  8 rs2989441_C       16     2.34
##  9 rs1561278_G       14     2.05
## 10 rs73212857_C      13     1.90
## # ... with 43 more rows
miss_var_table(data)
## # A tibble: 19 x 3
##    n_miss_in_var n_vars pct_vars
##            <int>  <int>    <dbl>
##  1             0     13    24.5 
##  2             1      1     1.89
##  3             2      6    11.3 
##  4             3      4     7.55
##  5             4      2     3.77
##  6             5      1     1.89
##  7             6      3     5.66
##  8             7      3     5.66
##  9             8      3     5.66
## 10             9      5     9.43
## 11            10      1     1.89
## 12            11      1     1.89
## 13            13      1     1.89
## 14            14      1     1.89
## 15            16      1     1.89
## 16            17      1     1.89
## 17            18      2     3.77
## 18            20      2     3.77
## 19            21      2     3.77
library(rpart)
library(rpart.plot)

data %>%
    add_prop_miss() %>%
    rpart(prop_miss_all ~ ., data = .) %>%
    prp(type = 4, extra = 101, prefix = "Prop. Miss = ")    
## Warning: Cannot retrieve the data used to build the model (so cannot determine roundint and is.binary for the variables).
## To silence this warning:
##     Call prp with roundint=FALSE,
##     or rebuild the rpart model with model=TRUE.

  1. 如图是影响缺失的主要变量。

由于缺失值不严重，因此进行lm。

library(broom)
## Warning: 程辑包'broom'是用R版本3.6.3 来建造的
data %>% 
    mutate_at(vars(-MPB)
              ,~fct_explicit_na(factor(.),'No_infos')) %>% 
    lm(MPB~.,data=.) %>% 
    tidy %>% 
    DT::datatable(
        rownames = FALSE,
        extensions = 'Buttons', options = list(
            dom = 'Bfrtip',
            buttons = c('copy', 'csv', 'excel', 'pdf', 'print')
      )
    )
  1. 点击对应格式可以下载。

Tierney, Nicholas. 2018a. “Exploring Imputed Values.” 2018. http://naniar.njtierney.com/articles/exploring-imputed-values.html.

———. 2018b. “Gallery of Missing Data Visualisations.” 2018. http://naniar.njtierney.com/articles/naniar-visualisation.html.

———. 2018c. “Getting Started with Naniar.” 2018. http://naniar.njtierney.com/articles/getting-started-w-naniar.html.