Look for date/time column in as.xts.data.frame()

[joshuaulrich]

It would be nice if as.xts.data.frame() didn't require date/times as row names, and instead looked for a time-based column in the data.frame if the row names weren't valid index values.

For example:

d <- structure(list(
    Date = structure(c(13515, 13516, 13517, 13518, 13519, 13520), class = "Date"),
    Open =  c(50.03, 50.23, 50.42, 50.37, 50.24, 50.13),
    High =  c(50.11, 50.42, 50.42, 50.37, 50.24, 50.21),
    Low =   c(49.95, 50.23, 50.26, 50.22, 50.11, 49.99),
    Close = c(50.11, 50.39, 50.33, 50.33, 50.18, 49.99)),
    row.names = c(NA, 6L), class = "data.frame")
as.xts(d)
##             Open  High   Low Close
## 2007-01-02 50.03 50.11 49.95 50.11
## 2007-01-03 50.23 50.42 50.23 50.39
## 2007-01-04 50.42 50.42 50.26 50.33
## 2007-01-05 50.37 50.37 50.22 50.33
## 2007-01-06 50.24 50.24 50.11 50.18
## 2007-01-07 50.13 50.21 49.99 49.99

[joshuaulrich]

@zeileis, @ggrothendieck: does zoo do anything like this currently? I know it would be harder because zoo doesn't have to be indexed by time. Maybe it could look for one column that was a different type than the rest in the data.frame?

[ggrothendieck]

read.zoo has an index= argument which defaults to 1 so for the example read.zoo(d) works. Most data sets have the index column first so there is really little advantage to automatically locating it.

[joshuaulrich]

Thanks for your thoughts! read.zoo() is actually one of the things that made me think of this (and good point about the index usually being in the first column).

I really like that read.zoo() provides this functionality. However, I don't think most users think of using a 'read' function to convert a data frame to a zoo/xts object. I imagine their first guess would be to use as.zoo() or as.xts(). That's my rationale for this enhancement.

[zeileis]

I agree with you. However, the as.zoo.data.frame() behaves like as.ts() on a data.frame. So we don't have any detection of potential index columns there. The question is whether it is worth changing the current behavior for this. What we could add in a fully backward-compatible way is:

as.zoo.data.frame(x, index = NULL, ...)

where index = NULL means that no index is provided explicitly (and 1, ..., n is used implicitly). This would give the users at least an option to do as.zoo(df, index = 1) but without any autodetection.

If we were willing to possibly break existing code we could interpret index = NULL as "please autodetect the index". Possible interpretations would be:

• Use the first column.
• Use the first column but only if it is not plain numeric or integer.
• Use the first non-plain numeric/integer column.
• Use the first and only non-plain numeric/integer column.

But I'm not sure if any of these really make life easier for existing zoo users. I would probably just add the argument without any auto-detection.

[ggrothendieck]

as.zoo.data.frame could be a wrapper around read.zoo with the appropriate default arguments to ensure backwards compatibility.