Clean way to enter formula method for types that don't require x (rect, segments)

[katrinabrock]

There may already be a nice way to do this, but I didn't see any examples actually using the formula method for rect and segments.

Using the example from type_segments:

df <- data.frame(x0 = c(0, .1), y0 = c(.2,1), x1 = c(1, .9
          ), y1 = c(.75, 0))

I would expect, I can do something like:

tinyplot(xmin=x0, ymin=y0, xmax=x1, ymax
          =y1, data = df, type = "segments")

This fails with x0 not found because, of course, I need xthe first argument to be a formula to get into the tinyplot.formula method.

Putting a placeholder formula does work:

tinyplot(y1~x1, xmin=x0, ymin=y0, xmax=x1, ymax=y1, data =
           df, type = "segments")

But this y1~x1 looks meaningful even though it isn't and therefore may confuse future readers of the code.

Here are some ideas that would make it more clear that the formula is a placeholder:

tinyplot(as.formula(NULL), xmin=x0, ymin=y0, xmax=x1, ymax
          =y1, data = df, type = "segments")

tinyplot:::tinyplot.formula(xmin=x0, ymin=y0, xmax=x1, ymax =y1, data = df, type = "segments")

tinyplot( ~1, xmin=x0, ymin=y0, xmax=x1, ymax=y1, data = df, type = "segments")

All three of these currently error out. Maybe there is some other clear formula placeholder already in the package that I haven't thought of, if so, it would be nice to document it more clearly and add it in an example to rect and segments help pages.

[grantmcdermott]

Thanks for raising and for the suggestions.

TBH, I don't have a great intuition about which of these alternatives makes the most sense. (Maybe @zeileis does?)

Another option that comes to mind is using . as a placeholder a la

tinyplot(.~., xmin=x0, ymin=y0, xmax=x1, ymax=y1, data = df, type = "segments")

... which is consistent with update.formula, but not necessarily with the way that the . is a placeholder for all vars in, say, lm(mpg ~ ., data = mtcars).

[grantmcdermott]

#347 might be related, in the sense that we want/require special logic for handling NULL x types. See https://github.com/grantmcdermott/tinyplot/pull/347/files

OTOH it might also just be that these cases are better handled through "direct" environment passing/scoping.

with(df, tinyplot(xmin=x0, ymin=y0, xmax=x1, ymax=y1, type = "segments"))

[katrinabrock]

Thanks for your thoughtful response!
That PR does seem adjacent. However, don't think the concerns raised in the PR about that particular change would apply to a change to resolve this case. as.formula(NULL), ~1, and .~. are all valid formulas so the changes needed would be restricted to the tinyplot.formula method. No new method or changes to the default method would be needed.

If it's recommended not to attempt using the formula method with types that don't require x, but rather add your own with( function, I think that should be noted in the documentation. Both in the helptext for the xmin,... arg and in the docs for the types that this applies to could be revised to clarify this.

[tony-aw]

In my humble opinion, I think it would be nice to have a tinyplot.data.frame method. Then you don't have to worry about the formula method in the described situations. And that way you can also safely use tinyplot_add() without relying on a formula, provided the same data.frame is used.

[zeileis]

Thanks for the original suggestion. Supporting the formula ~ 1 as a specification for x = NULL, y = NULL is a good idea and seems straightforward to implement. I'm happy to try this.

The "formula" as.formula(NULL) is not a valid formula. It has the class "formula" but it cannot be processed by terms(). I'll have another look at why this happens, looks like a bug in as.formula() to me.

The formula . ~ . has a clear meaning and in my opinion it doesn't fit the context of this application.

Re: tinyplot.data.frame...I'm still uncertain about that one.

[grantmcdermott]

@tony-aw I agree it would be cool, but there are some legitimate concerns about unintended consequences that we’re able to resolve just yet. So it’s on ice for the moment.

@zeileis cool. You’re our resident formula expert, so I’ll defer to your judgement. Thanks for the taking a look and possible PR when you can 👍

[tony-aw]

May I ask what concerns you have about a data.frame based plotting mechanism?
It doesnt have to be tinyplot.data.frame.

[katrinabrock]

@tony-aw discussion of the data.frame method is taking place on #347 . The issue that PR is trying to address is related to, but not the same as this one. Not sure what you mean by a data.frame based plotting mechanism that is not a tinyplot.data.frame method. I would argue that the existing formula method is already "data.frame based" in some sense.