oce.plot.ts() fails with "need finite 'ylim' values" even though plot() works

[dankelley]

The dataset is from a glider and so creating a regexp is a bit tricky. I plan to instrument the code and to see where the problem lies. Then I may be able to see how to make a regexp, and I'll insert that (possibly with a better titles).

[dankelley]

The problem is that it is trying to simplify the vectors, which exceed the limit. I guess when it does that, it is not happening upon finite values. See below (not for others reading this -- just my notes to myself).

> table(is.finite(S))

FALSE  TRUE 
12028  2447 
> i <- 1:100; oce.plot.ts(t[i],is.finite(S[i]), type = "p")

[dankelley]

I'll look into this tomorrow. That bit of the code is a bit on the tricky side. But I like knowing about this issue, because it has not come up in my usual workflow (where there are normally more finite values than NA ones; here there are 6 NA values per non-NA value. Glider data are fun :-)

[dankelley]

It now checks for the condition. If ylim is given, then a plot is shown but there are no data, and a message about no non-NA y values is written in the plot region. But if ylim is not given, there is no way for it to proceed, so it says as in the example below. This is better than telling users that there are no non-NA values (which is what plot() does).

Tomorrow I'll look into seeing why simplification does this. There may be a way to retain some data. But I need to think through what makes most sense, in the "R way of thinking", i.e. what users would expect.

Error in oce.plot.ts(t, S, debug = 0) : 
  simplification eliminated all y data; try using simplify=NA

[dankelley]

The problem is that the data just happen to be sparser than I had thought could occur. The way simplification works for type="l" is that it splits the data up into subsets, and then computes the min and max of y in each of those, to draw a vertical line at that time value. But the code is doing that only if there is more than 1 value in the subset; otherwise it uses NA.

If there is only one value there, then the min will equal the max and so the vertical line will have zero extent. But maybe that's OK. At least it seems OK to me right now. I need to mess with some sample datasets and also to scratch my head for a while.

Maybe sometime if @richardsc and I are in a room together, we can chat about this, since simplification was very much a jointly -developed idea between the two of us.

[dankelley]

Here is what I get on my problematic case of (very sparse) glider data, with not-yet-push-to-GH code.