prompt on unavailable extension packages

[simonpcouch]

Closes #731.

I've had a few moments of confusion as I've familiarized with parsnip where I unknowingly specify a model specification/engine/mode combination that requires an extension package, and only come across failures later on in a modeling pipeline. This seems to be the source of confusion in #544 as well.

This PR refines print methods to nudge users to load an extension package if it's needed.

# warns informatively when it ought to:
library(parsnip)

C5_rules()
#> parsnip could not locate an implementation for `C5_rules` classification model specifications using the `C5.0` engine.
#> ℹ The parsnip extension package rules implements support for this specification.
#> ℹ Please install (if needed) and load to continue.
#> 
#> C5.0 Model Specification (classification)
#> 
#> Computational engine: C5.0

C5_rules(engine = "C5.0") %>% translate()
#> parsnip could not locate an implementation for `C5_rules` classification model specifications using the `C5.0` engine.
#> ℹ The parsnip extension package rules implements support for this specification.
#> ℹ Please install (if needed) and load to continue.
#> 
#> C5.0 Model Specification (classification)
#> 
#> Computational engine: C5.0

# doesn't prompt and prints out the call as before if the implementing extension is loaded:
library(rules)

C5_rules()
#> C5.0 Model Specification (classification)
#> 
#> Computational engine: C5.0

C5_rules(engine = "C5.0") %>% translate()
#> C5.0 Model Specification (classification)
#> 
#> Computational engine: C5.0 
#> 
#> Model fit template:
#> rules::c5_fit(x = missing_arg(), y = missing_arg(), weights = missing_arg())

# otherwise behaves the same as before:
linear_reg() %>% translate()
#> Linear Regression Model Specification (regression)
#> 
#> Computational engine: lm 
#> 
#> Model fit template:
#> stats::lm(formula = missing_arg(), data = missing_arg(), weights = missing_arg())

linear_reg()
#> Linear Regression Model Specification (regression)
#> 
#> Computational engine: lm

^{Created on 2022-05-27 by the reprex package (v2.0.1)}

[EDIT: updated reprex and explanation to reflect changes re: Julia's edits]

[juliasilge]

I think this would make the most sense in an rlang::inform() vs. printing (and we have a lot of tools for handling the rlang messages).

[simonpcouch]

Ah, thank you @juliasilge! I agree that rlang::inform likely makes more sense.

Just pushed those changes—they're a bit more invasive now but do result in a better interface.

[simonpcouch]

Wrapping up for the night, though this isn't quite there.🌝🌚

[simonpcouch]

A quick self-review to give some context for changes. :)

[juliasilge]

It will be a huge help to give more useful info in this situation! 🙌

I've got a bit of specific feedback and then some questions about the approach.

[juliasilge]

Is this an expensive check to do every time new_model_spec() is run (which is run a lot, I think)? The env is just there, but what about reading the file?

[simonpcouch]

Certainly more expensive than our other checks. But—

• It’s only run on initiation of new model types (i.e. never at set_engine, update, etc)
• It’s possibly negligible compared to fit times:

library(parsnip)

# running the check
system.time(mod <- C5_rules())
#> parsnip could not locate an implementation for `C5_rules` classification model specifications using the `C5.0` engine.
#> ℹ The parsnip extension package rules implements support for this specification.
#> ℹ Please install (if needed) and load to continue.
#>    user  system elapsed 
#>   0.038   0.001   0.039

# doesn't run the check
system.time(mod <- mod %>% set_engine("C5.0"))
#>    user  system elapsed 
#>   0.008   0.000   0.009

# model fit to a lil fella
library(rules)

mtcars$cyl <- as.factor(mtcars$cyl)

system.time(mod %>% fit(cyl ~ ., mtcars))
#>    user  system elapsed 
#>   0.445   0.020   0.467

^{Created on 2022-05-25 by the reprex package (v2.0.1)}

[juliasilge]

The fact that it could be common to read this file multiple times seems like something to check and get other feedback on. We could read this file one time at .onLoad or something? Let's have one more person look at this part of the PR and see what they think.

[simonpcouch]

I think that's a good observation. After 9918043, parsnip now assigns the models.tsv system file as model_info_table internally so we can skip that load step on checks:

library(parsnip)
  
# now just assigns models.tsv to internal object:
head(parsnip:::model_info_table)
#>      model                mode engine      pkg
#> 1 bag_mars      classification  earth baguette
#> 2 bag_mars          regression  earth baguette
#> 3 bag_tree censored regression  rpart censored
#> 4 bag_tree      classification   C5.0 baguette
#> 5 bag_tree      classification  rpart baguette
#> 6 bag_tree          regression  rpart baguette

# running the check. used to be:
#>    user  system elapsed 
#>   0.038   0.001   0.039
system.time(mod <- C5_rules())
#> parsnip could not locate an implementation for `C5_rules` classification model specifications using the `C5.0` engine.
#> ℹ The parsnip extension package rules implements support for this specification.
#> ℹ Please install (if needed) and load to continue.
#>    user  system elapsed 
#>   0.033   0.001   0.033


# doesn't run the check. used to be:
#>    user  system elapsed 
#>   0.008   0.000   0.009
system.time(mod <- mod %>% set_engine("C5.0"))
#>    user  system elapsed 
#>   0.007   0.000   0.007


# model fit to a lil fella, for comparison
library(rules)

mtcars$cyl <- as.factor(mtcars$cyl)

system.time(mod %>% fit(cyl ~ ., mtcars))
#>    user  system elapsed 
#>   0.440   0.025   0.464

^{Created on 2022-05-27 by the reprex package (v2.0.1)}

I opted not to ditch that models.tsv file entirely as it seems our machinery to keep it updated with the current docs system works well.

[juliasilge]

The fact that it could be common to read this file multiple times seems like something to check and get other feedback on. We could read this file one time at .onLoad or something? Let's have one more person look at this part of the PR and see what they think.

[simonpcouch]

This should be ready for your review, @topepo!

One unresolved discussion that we'd especially appreciate your eye on: #732 (comment)

[simonpcouch]

Note from meeting: thumbs up from Max to merge!

[github-actions]

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.