NA-row Handling on timeseries functions

[lawinslow]

We have "na.rm" option, but what happens if an entire row is NA? (#43) If we skip such a row, we will still be returning an NA, breaking the API contract.

Should we drop that row? Should we return an NA? Should we throw an error?

[jzwart]

You mean the user has the expectation they will not have any NA's returned to them if they use na.rm?

I think it should still return NA if an entire row is NA; a warning might be helpful to the user or modify the documentation to indicate that the na.rm option makes the tool calculate given physical parameter based on non-NA data, if all data is NA, then it returns an NA.

If it removes a row, then the data.frame dimensions get changed - I find that annoying when that happens.

[lawinslow]

Interesting

sum(c(NA,NA,NA), na.rm=TRUE)
[1] 0

[lawinslow]

max(c(NA,NA,NA), na.rm=TRUE)
[1] -Inf
Warning message:
In max(c(NA, NA, NA), na.rm = TRUE) :
  no non-missing arguments to max; returning -Inf

[jzwart]

Huh. This is almost getting philosophical... the sum of nothing is 0 but the max of nothing is -Inf.

I like the way max() returns a warning message

[lawinslow]

Sometimes when defining APIs, it gets philosophical. I just wanted to know what base functions were doing with this challenge. Clearly, it's kinda weird behavior that I don't think we'd want to adopt (return -Inf or 0, nope).

Let's call na.rm "best effort". Warning if it can't handle the input well, but still return an NA. I agree that it is fairly nice if the output is of same nrow as input.

I propose applying this style of handling to all other timeseries functions.

[jzwart]

That sounds good to me.

On Thu, Dec 4, 2014 at 5:02 PM, Luke Winslow [email protected]
wrote:

Sometimes when defining APIs, it gets philosophical. I just wanted to know
what base functions were doing with this challenge. Clearly, it's kinda
weird behavior that I don't think we'd want to adopt (return -Inf or 0,
nope).

Let's call na.rm "best effort". Warning if it can't handle the input
well, but still return an NA. I agree that it is fairly nice if the output
is of same nrow as input.

I propose applying this style of handling to all other timeseries
functions.

—
Reply to this email directly or view it on GitHub
#45 (comment).

Jacob A. Zwart
University of Notre Dame
Jones Lab & MFE
263 Galvin Life Sciences
Notre Dame, IN 46556
269.370.2788