Release schedule

[vincentarelbundock]

Sorry to bug you with this, but I'm curious to know if there's a plan for a next release in the near future (I see that they are quite frequent on the archive page).

I don't mean to put pressure at all. Just thinking about coordinating a release with insight dependency and trying to figure out which version I can require.

Again, no pressure! I know releasing to CRAN can be a pain.

[bwiernik]

Follow the "dataverse cycle" issue in easystats for where we are in the release cycles

[vincentarelbundock]

Got it, thanks!

easystats/easystats#143

[strengejacke]

We have easystats::on_CRAN(), and I usually try to avoid submitting packages in red. That said, we could plan a new release soon:

What is your schedule for the next releases of your packages? I think it should be fine to coordinate these with insight, and submitting to CRAN is usually quite smooth, unless you have some breaking changes.

[vincentarelbundock]

Ah, the on_cran function is very useful, thanks!

I received an ominous email from Ripley saying that I need to release a new version before October 19th.

(One of my tests fails at tolerance 0.000001 on the new ARM mac. Catastrophe!)

[vincentarelbundock]

@strengejacke I'm not sure what your process is for updating NEWS.md. But FYI, I skimmed through the commits since last release and added some stuff there.

[vincentarelbundock]

6822362

[bwiernik]

Thanks! Our process is "not as organized as it should be"

[vincentarelbundock]

I chuckled.

[strengejacke]

Any critical issues that need to be addressed before submitting to CRAN? Else, I might submit an insight update this week.

[bwiernik]

Maybe the glmmTMB fixes Vincent was working on?

[vincentarelbundock]

My guess is that the glmmTMB fixes are going to require a little bit of infrastructure. I can take a look at that eventually, but my next couple weeks are going to be busy at work, so I'm unlikely to have as much time for hobby coding.

In any case, this only applies to a small class models, and we wouldn't be introducing new bugs. So I suggest we raise a warning for zero-inflated models when predict="expectation", and we take a bit more time to figure out what the most generalizable/solid infrastructure is going to be for glmmTMB and similar cases.

[Disclosure: I am certainly biased, since CRAN is going to remove marginaleffects if I don't update it, and I made the mistake of developing against insight devel features, so current master is all sorts of broken against 0.14.4.]

[vincentarelbundock]

Aside from glmmTMB, I don't see other major issues, and I have no immediate plans for new features.

[vincentarelbundock]

FYI, I added the proposed warning. I can be removed when we implement the zero-inflation correction: dca9e72

[strengejacke]

[Disclosure: I am certainly biased, since CRAN is going to remove marginaleffects if I don't update it, and I made the mistake of developing against insight devel features, so current master is all sorts of broken against 0.14.4.]

Great, you're now following @IndrajeetPatil footsteps. CRAN team already knows that every new submission of any easystats package will 99% sure break any of @IndrajeetPatil packages ;-)

[strengejacke]

Maybe we can add easystats/modelbased#151 (comment). I think this should be easy, and the regex stuff to detect transformations can be copied from https://github.com/strengejacke/ggeffects/blob/master/R/utils_ggpredict.R

[strengejacke]

Does anybody know, if I have a string "log(x+1)", how can I create a function that does function(x) log(x + 1)? Or when I have "log(x+5)", I would like to return a function function(x) log(x + 5).

[IndrajeetPatil]

Something like this?

arg <- quote(x + 1)
rlang::call2("log", arg)
#> log(x + 1)

arg <- quote(x + 5)
rlang::call2("log", arg)
#> log(x + 5)

^{Created on 2021-10-15 by the reprex package (v2.0.1)}

[bwiernik]

Maybe str2lang() ?

[strengejacke]

Nope.

fun <- function(x) str2lang("log(x+2)")

# is
fun(.3)
#> log(x + 2)

# should be
log(.3 + 2)
#> [1] 0.8329091

arg <- quote(x + 1)
fun <- function(x) rlang::call2("log", arg)
fun(.3)
#> log(x + 1)

^{Created on 2021-10-15 by the reprex package (v2.0.1)}

[IndrajeetPatil]

Here is a working example:

library(rlang)

foo <- function(x, a = 1) {
  args <- list(x = quote(x + a))
  eval(call2(.fn = log, !!!args))
}

foo(0.3)
#> [1] 0.2623643

log(0.3 + 1)
#> [1] 0.2623643

foo(0.3, 2)
#> [1] 0.8329091

log(0.3 + 2)
#> [1] 0.8329091

^{Created on 2021-10-15 by the reprex package (v2.0.1)}

[strengejacke]

What I exactly want:

1. We have lm(log(y+2) ~x, data = d)
2. The function get_transformation() extracts the response transformation and extracts the string "log(x+2)".
3. Now get_transformation() should return a function foo(x), that internally does exact the same thing as defined in the above string (i.e. log(x+2).

[IndrajeetPatil]

So, if I understand correctly, you want a function such that if I enter a string "log(x+2)" into it as an argument, it outputs a function of the form: function(x) log(x + 2), and not a language type object (log(x + 2))?

[strengejacke]

yes. Just that you don't enter a string, but it's rather extracted from the model object. See these new two functions:

https://github.com/easystats/insight/blob/master/R/find_transformation.R
https://github.com/easystats/insight/blob/master/R/get_transformation.R

[strengejacke]

For any lm(log(y+ <number>) ~ x, data = d), find_transformation() returns "log(x+<number>)". And get_transformation() should return a function that accepts one argument, and then applies the transformation that was used in the formula.

[strengejacke]

The idea came from here: easystats/modelbased#151 (comment)

[strengejacke]

@vincentarelbundock @bwiernik @IndrajeetPatil

Ok, I'm running final checks now and will submit insight today. I think we can add more features to get_transformation() later, the most important stuff should work for now.

[vincentarelbundock]

That's fantastic! Thanks so much!

[vincentarelbundock]

wait, might be an issue in insight, actually.

On the one hand, rev deps are kind of a pain, but on the other hand it's kind of neat that we can benefit from the test suites of downstream packages. 3ae2bd2 is another example of a minor bug that I had introduced in prior commit and which I discovered because of marginaleffects' tests

[strengejacke]

on CRAN :-)

[vincentarelbundock]

on CRAN :-)

Yessss