Plotting inside function: subset(df,id_==…) gives

问题:

I have a df with multiple y-series which I want to plot individually, so I wrote a fn that selects one particular series, assigns to a local variable dat, then plots it. However ggplot/geom_step when called inside the fn doesn't treat it properly like a single series. I don't see how this can be a scoping issue, since if dat wasn't visible, surely ggplot would fail?

You can verify the code is correct when executed from the toplevel environment, but not inside the function. This is not a duplicate question. I understand the problem (this is a recurring issue with ggplot), but I've read all the other answers; this is not a duplicate and they do not give the solution.

set.seed(1234)
require(ggplot2)
require(scales)

N = 10
df <- data.frame(x = 1:N,
                 id_ = c(rep(20,N), rep(25,N), rep(33,N)),
                 y = c(runif(N, 1.2e6, 2.9e6), runif(N, 5.8e5, 8.9e5) ,runif(N, 2.4e5, 3.3e5)),
                 row.names=NULL)

plot_series <- function(id_, envir=environment()) {
  dat <- subset(df,id_==id_)
  p <- ggplot(data=dat, mapping=aes(x,y), color='red') + geom_step()
  # Unsuccessfully trying the approach from http://stackoverflow.com/questions/22287498/scoping-of-variables-in-aes-inside-a-function-in-ggplot
  p$plot_env <- envir
  plot(p)
  # Displays wrongly whether we do the plot here inside fn, or return the object to parent environment 
  return(p)
}

 # BAD: doesn't plot geom_step!
plot_series(20)

# GOOD! but what's causing the difference?
ggplot(data=subset(df,id_==20), mapping=aes(x,y), color='red') + geom_step()

#plot_series(25)
#plot_series(33)

回答1:

This works fine:

plot_series <- function(id_) {
    dat <- df[df$id_ == id_,]
    p <- ggplot(data=dat, mapping=aes(x,y), color='red') + geom_step()
    return(p)
}

print(plot_series(20))

If you simply step through the original function using debug, you'll quickly see that the subset line did not actually subset the data frame at all: it returned all rows!

Why? Because subset uses non-standard evaluation and you used the same name for both the column name and the function argument. As jlhoward demonstrates above, it would have worked (but probably not been advisable) to have simply used different names for the two.

The reason is that subset evaluates with the data frame first. So all it sees in the logical expression is the always true id_ == id_ within that data frame.

One way to think about it is to play dumb (like a computer) and ask yourself when presented with the condition id_ == id_ how do you know what exactly each symbol refers to. It's ambiguous, and subset makes a consistent choice: use what's in the data frame.

回答2:

Notwithstanding the comments, this works:

plot_series <- function(z, envir=environment()) {
  dat <- subset(df,id_==z)
  p <- ggplot(data=dat, mapping=aes(x,y), color='red') + geom_step()
  p$plot_env <- envir
  plot(p)
  # Displays wrongly whether we do the plot here inside fn, or return the object to parent environment 
  return(p)
}

plot_series(20)

The problem seems to be the subset is interpreting id_ on the RHS of the == as identical to the LHS, to this is equivalent to subletting on T, which of course includes all the rows of df. That's the plot you are seeing.