'skip' in createTimeSlices()

[praftery]

Minor request here.

1. The 'skip' arg for createTimeSlices is not available when passing trainControl(method = "timeslice"...) to train(), which means a user must create the time slices themselves if they need to reduce the sample space. That's a very common need for timeseries analysis. For example, if doing a timeseries analysis with 1 minute data over a one month period, you can easily end up with thousands of train/test samples...

2. The 'skip' parameter itself is not that useful, as a user has to do some math to figure out just how many samples will be generated when creating the slices. It might be more useful to specify the total numer of test samples (n_samples), and randomly select them without replacement from the possible samples.

i.e. replace line 25 in createTimeSlices.R:
x[seq(1, n, by = skip)]
with
sample(x, n_samples, replace = FALSE)

[topepo]

I can add it to the control function.

2. The 'skip' parameter itself is not that useful

We're going to disagree on that. It is a little math but it makes sense for week and year data (e.g. skip = 6 when you want the same day across weeks).

> library(lubridate)
> dates <- today() + days(1:17)
> 
> ind <- createTimeSlices(1:length(dates), initialWindow = 3, skip = 6)
> lapply(ind$test, function(x, y) wday(ymd(y[x])), y = dates)
$Testing01
[1] 6

$Testing08
[1] 6

Using sample is not a good option since you probably want to equally spaced gaps.

[praftery]

I see your point about random sampling vs. skip. Thanks for building in this feature - it's very useful for timeseries crossvalidation, and I appreciate it.