let Example

Nina Zumel

2018-07-20

An example of using let to wrap dplyr expressions as functions.

library("dplyr")
library("replyr")

The desired task: write a function that takes a data frame with a specified numerical column and an optional grouping column, and returns a data frame with one row per group containing:

The dplyr expression for such a table is easy when the column names are known, but complicated when they are not. We use wrapr::let to write such a function without the use of lazyeval or rlang/tidyeval.

sumstat_intervals = function(dframe, colname, groupcolname = NULL) {
  mapping = list(COLNAME  =colname,
                 GROUPCOLNAME = groupcolname)
  let(alias = mapping,
      {
        if(!is.null(groupcolname)) {
          dframe <- group_by(dframe, GROUPCOLNAME)
        }
        summarize(dframe, 
                  sdlower = mean(COLNAME)-sd(COLNAME),
                  mean = mean(COLNAME),
                  sdupper = mean(COLNAME) + sd(COLNAME),
                  iqrlower = median(COLNAME)-0.5*IQR(COLNAME),
                  median = median(COLNAME),
                  iqrupper = median(COLNAME)+0.5*IQR(COLNAME))
      })
}

We can test sumstat_intervals on iris:

sumstat_intervals(iris, "Sepal.Length")
 #     sdlower     mean  sdupper iqrlower median iqrupper
 #  1 5.015267 5.843333 6.671399     5.15    5.8     6.45
sumstat_intervals(iris, "Sepal.Length", "Species")
 #  # A tibble: 3 x 7
 #    Species    sdlower  mean sdupper iqrlower median iqrupper
 #    <fct>        <dbl> <dbl>   <dbl>    <dbl>  <dbl>    <dbl>
 #  1 setosa        4.65  5.01    5.36     4.8     5       5.2 
 #  2 versicolor    5.42  5.94    6.45     5.55    5.9     6.25
 #  3 virginica     5.95  6.59    7.22     6.16    6.5     6.84
sumstat_intervals(iris, "Petal.Length", "Species")
 #  # A tibble: 3 x 7
 #    Species    sdlower  mean sdupper iqrlower median iqrupper
 #    <fct>        <dbl> <dbl>   <dbl>    <dbl>  <dbl>    <dbl>
 #  1 setosa        1.29  1.46    1.64     1.41   1.5      1.59
 #  2 versicolor    3.79  4.26    4.73     4.05   4.35     4.65
 #  3 virginica     5.00  5.55    6.10     5.16   5.55     5.94

We can also use let to parameterize other functions that specify their parameters via non-standard evaluation. For example, we could write a ggplot2 function to plot the information in sumstat_intervals (either the mean-centered interval or the median-centered one) using ggplot2::aes_string. Or we could use wrapr::let.

plot_distributions = NULL

if  (requireNamespace("ggplot2")) {
  library("ggplot2")
  plot_distributions = function(dframe, colname, groupcol,
                                intervaltype="mean", title="") {
    if(!(intervaltype %in% c("mean", "median")))
      error("Intervaltype must be one of 'mean' or 'median'")
    
    sintervals = sumstat_intervals(dframe, colname, groupcol)
    
    # I could do the following with aes_string, but what the heck
    mapping = list(xval=groupcol, yval=colname, center=intervaltype)
    if(intervaltype=="mean") {
      mapping2 =list(lower="sdlower", upper="sdupper")
    } else {
      mapping2 =list(lower="iqrlower", upper="iqrupper")
    }
    mapping = c(mapping, mapping2)
    
    let(alias=mapping,
        expr = {
          ggplot(dframe, aes(x=xval,color=xval)) +
            geom_jitter(aes(y=yval), width=0.2, height=0, alpha=0.5) +
            geom_crossbar(data=sintervals, aes(y=center, ymin=lower, ymax=upper)) +
            ggtitle(title) + theme(plot.title=element_text(hjust=0.5)) +
            scale_color_brewer(palette="Dark2")
        })
  }
}
 #  Loading required namespace: ggplot2
if(!("NULL") %in% class(plot_distributions)) {
  plot_distributions(iris, "Sepal.Length", "Species",
                     title="Iris sepal length with mean +/1 one standard deviation")
}

if(!("NULL") %in% class(plot_distributions)) {
  plot_distributions(iris, "Petal.Width", "Species",
                     intervaltype="median",
                     title="Iris petal width with median and centered IQR interval")
}