`type = "jitter"` does not work with date on x-axis

[wachtermh]

Consider the following example:

library(tinyplot)

data("airquality")

airquality$Year <- 1973
airquality$DateChr <- paste0(airquality$Year, "-", airquality$Month)  
airquality$Date <- as.Date(paste0(airquality$DateChr, "-01"), format = "%Y-%m-%d")

Doing

plt(Temp ~ DateChr, data = airquality, type = "jitter")

works:

However doing

plt(Temp ~ Date, data = airquality, type = "jitter")

does not work and gives Error in jitter(x, factor = factor, amount = amount) : 'x' must be numeric.

Maybe something like this could work:

data_jitter = function(factor, amount) {
  fun = function(datapoints, ...) {
    x = datapoints$x
    y = datapoints$y
    if (is.factor(x)) {
      xlvls = levels(x)
      xlabs = seq_along(xlvls)
      names(xlabs) = xlvls
      x = as.integer(x)
    } else if (inherits(x, c("Date", "POSIXt"))) {
      xfactor = factor(x, unique(x))
      xlvls = levels(xfactor)
      xlabs = seq_along(xlvls)
      names(xlabs) = xlvls
      x = as.integer(xfactor)
    } else {
      xlabs = NULL
    }
    if (is.factor(y)) {
      ylvls = levels(y)
      ylabs = seq_along(ylvls)
      names(ylabs) = ylvls
      y = as.integer(y)
    } else {
      ylabs = NULL
    }
    x = jitter(x, factor = factor, amount = amount)
    y = jitter(y, factor = factor, amount = amount)
    
    datapoints$x = x
    datapoints$y = y
    
    out = list(
      datapoints = datapoints,
      x = x,
      y = y,
      xlabs = xlabs,
      ylabs = ylabs
    )
    return(out)
  }
}

plt(Temp ~ Date, data = airquality, type = type_jitter())

[zeileis]

Thanks for the report! The reason for this is that we have an explicit workaround for factor variables to which character variables get coerced - but we don't have a workaround for date and date/time objects (like Date, POSIXct, yearmon, etc.).

The (non-generic) function jitter() does not work for these objects because they are not numeric although they are actually just numeric vectors plus a class and potentially further attributes.

d <- as.Date("2000-01-01")
jitter(d)
## Error in jitter(d) : 'x' must be numeric

A quick & dirty workaround could be to drop the class and just add it again later...at the risk that (a) this still does not work and the error message might not be more intelligible and (b) the resulting object might not be strictly valid.

jitter2 = function(x, factor = 1, amount = NULL) {
  y = jitter(unclass(x), factor = factor, amount = amount)
  class(y) = class(x)
  return(y)
}

Using this works correctly for Date and POSIXct objects and does not fail horribly for yearmon or yearqtr objects. However, the xlim might be a bit off because the jittered objects are not strictly on a grid of 1/12 or 1/4 steps anymore which leads to xlim values that are a bit off.

[wachtermh]

Thank your for the answer! I wonder if this could be avoided by setting xlim for type_jitter to something like this:

xlim = range(pretty(range(jitter2(x))))

[zeileis]

For Date and POSIXct this seems to be unnecessary and for yearmon and yearqtr it is just less bad. 😅

[grantmcdermott]

Hmmm. @zeileis do you think this requires a proposed upstream fix so that jitter does become a generic? Or is that going to be too slow and problematic?

[zeileis]

I don't have high hopes that jitter would become generic. And even if we can convince R Core to introduce the generic it would take even longer until contributed packages would pick it up for their classes.

Hence I would probably do something quick & dirty for now that works for standard classes like Date and POSIXct and has the chance to work "not terribly" for similar classes. And then I would wait to see who complains 🙈

[vincentarelbundock]

See #544 for "something quick & dirty" 😄