NA values break plot.errors

[richardsc]

Just discovered this package (after thinking about writing my own) -- it's great! Thanks!

I noticed that xlim and ylim are hard-coded in the NextMethod() call for plot.errors() to be range(x) and range(y).

One problem with this is that it is impossible to control the axis limits of the plot (say, to focus on a region or reverse an axes).

Another issue is that if x or y have NAs they cause plot.errors() to fail:

library(errors)
x <- 1:100
y <- rnorm(x)
y[20] <- NA
errors(x) <- 0
errors(y) <- rnorm(y, sd=0.25)
plot(x, y)

Error in plot.window(...) : need finite 'ylim' values

I don't see why plot.errors() isn't simply written as:

plot.errors <- function(x, y, ...) {
  if (missing(y)) {
    y <- x
    x <- seq_along(x)
  }

  NextMethod()

  if (inherits(x, "errors"))
    graphics::segments(errors_min(x), y, errors_max(x), y)

  if (inherits(y, "errors"))
    graphics::segments(x, errors_min(y), x, errors_max(y))
}

Note that using the na.rm=TRUE argument in range() doesn't help, as it still returns c(NA, NA) for range(y). I assume that this is due to the interaction the base range() function with the errors-class object and it doesn't know what to do with it?

[dankelley]

I also have come to the errors package only in the past few weeks (and I had actually written a similar, but much less useful package, a few days ago).

This range problem is in R/summary.R line 32, viz.

    "range" = c(min(...), max(...))

which I think ought to read

    "range" = c(min(..., na.rm=na.rm), max(..., na.rm=na.rm))

[dankelley]

Oh, and also the which.min and which.max calls above line 32 could also use adjustment.

[Enchufa2]

Thanks for your reports.

@dankelley You are right about range, it should be fixed in 8f95673. which.min and which.max have no na.rm argument though.

@richardsc

I don't see why plot.errors() isn't simply written as [...]

The problem is that we don't want error bars to be cut out of the plotting range by default, but I agree that a user-provided xlim or ylim should be respected. So it's a little more complicated than that. I'm working on it.

[Enchufa2]

In fact, it was a incorrect implementation of min, max and range. Rewritten in 61b004c.

[Enchufa2]

plot fixed in a3a6f0f, I think. Could you please test the current master branch, @richardsc?