Updates to slab()

[brownag]

These are some updates that close #229

• slab() now uses dice() so users will not get confused when they get a deprecation message RE: slice() while using slab(). I have had several questions from users about this (essentially: how do I use dice() with slab()?)

library(aqp)
#> This is aqp 1.43
data(sp1, package = 'aqp')
depths(sp1) <- id ~ top + bottom
a.max <- slab(sp1, fm = ~ prop)
#> Note: aqp::slice() will be deprecated in aqp version 2.0
#> --> Please consider using the more efficient aqp::dice()

• fixes mismatching lengths of slab labels v.s rows in data; allows for overlap; see .genSlabLabels2()

• added and exported slab_function() a helper method for accessing the internal slab methods provided by the package; now used in method definitions rather than internal functions

  • add more documentation on arguments/patterns for custom functions, and ensuring extra arguments (or lack thereof)) work robustly

• added a weighted variant of the numeric/quantile method via Hmisc::wtd.quantile(), and support for custom weighted functions (such as weighted.mean())

• weighted variant of categorical method (created stub function .slab.fun.factor.weighted()) now in slab(): weighted variant of categorical method  #269

[dylanbeaudette]

Looks good, please have a look at specific comments. I would like to do some testing before fully converting the unweighted stuff to DT.

[brownag]

I had mentioned I wanted to check something with the different possible ways of specifying slab.structure. It appears that everything is working as expected. I will re-request a review, no rush on checking it over,

[dylanbeaudette]

I thought of one potential issue, found last night while working on NCSP and dice: when using a formula that specifies all variables ~ . the horizon metadata variables will be included in the calculation of percent missing. This isn't usually a problem, but something to keep in mind as we move forward. slab always includes a named list of variables on the formula RHS, so this isn't a problem. We just need to be consistent in how we carry forward "horizon metadata", as a future plan for slab includes output as an SPC.

[brownag]

I thought of one potential issue, found last night while working on NCSP and dice: when using a formula that specifies all variables ~ . the horizon metadata variables will be included in the calculation of percent missing. This isn't usually a problem, but something to keep in mind as we move forward.

It's funny you mention this, since given the discussion we had on dice() with that type of formula I considered this in testing

slab always includes a named list of variables on the formula RHS, so this isn't a problem.

Yep. I think it would be cool if slab() could support the ~ . syntax.

We just need to be consistent in how we carry forward "horizon metadata", as a future plan for slab includes output as an SPC.

Yeah, agreed. Would like to see a method for converting slab output to SPC, I actually was going to draft one... and if we were to set that up... then we of course would need to address the metadata issue for consistency in "SPC methods that return SPCs"

[brownag]

Assuming I navigated the parallel changes to master branch with the new "factor" check dataset correctly-- I find no difference between the v1 and v2 slab methods when using ~ genhz formula.

[brownag]

I've looked over the KSSL check code and made updates last weekend, the numeric discrepancy for that check set is in the Musick profiles that don't pass a logic check.

Remaining comments and questions:

• Are the "changes" in value (generally small errors that result in very similar value after rounding) and contributing fraction acceptable? These differences only occur in the case of profiles that have invalid depth logic to begin with.
• Is there any issue with contributing fraction calculation in new code? Are we properly accounting for overlaps and gaps that are now allowed as inputs?

[brownag]

From my testing I find that the cpm=2 argument also gives the same results for ~genhz formula between version 1 and 2.