Adding tidiers for regression models not currently covered by `broom`

[IndrajeetPatil]

Here are some commonly used (class of) regression models that broom currently doesn't support but would be lovely if it does.

• zero-inflated regression: zeroinfl (https://cran.r-project.org/web/packages/pscl/vignettes/countreg.pdf)

• hurdle models for count data regression: hurdle (https://cran.r-project.org/web/packages/pscl/vignettes/countreg.pdf)

• robust regression from robustbase package: lmrob (https://cran.r-project.org/web/packages/robustbase/robustbase.pdf)

• multivariate ordinal regression models: mvord (https://cran.r-project.org/web/packages/mvord/vignettes/vignette_mvord2.html)

• multinomial logit model: mnlogit (https://cran.r-project.org/web/packages/mnlogit/index.html)

• vector generalized linear models: vglm (https://cran.r-project.org/web/packages/VGAM/VGAM.pdf)

• outlier-robust fit for Generalized Additive Models: rgam (https://cran.r-project.org/web/packages/rgam/index.html)

• resistant regression: lqs (https://www.rdocumentation.org/packages/MASS/versions/7.3-51.4/topics/lqs)

• Robust Linear Regression with Compositional Data as Covariates : complmrob (https://cran.r-project.org/web/packages/complmrob/complmrob.pdf)

(I will keep updating this list as I find new, unsupported regression models.)

Unless somebody else does it first, I will also try to do a PR with these tidiers.

[dgrtwo]

Interesting: we have implementations for lmRob and glmRob from the robust package (though they're basically thin wrappers for the lm tidiers), but I hadn't heard of the lowercase versions. Sounds useful!

[bbolker]

when you (or someone, maybe me?) does pscl::zeroinfl, (a) worth doing pscl::hurdle at the same time, (b) worth trying to maintain consistency with the conventions in broom.mixed, which already has some zero-inflation support for glmmTMB models ...

[billdenney]

What about tidying the output of optim()? A minor challenge with this is that it returns a list, not a named object class (so, the tidy method would need to look at the list and see if it appeared to be optim() output).

If interested, I'd be happy to generate a PR.

[alexpghayes]

There currently are tidiers for optim()! Do these work well for you, or do you have suggested changes?

library(broom)

r <- function(x) (3 - x[1])^2 + 4 * (x[2] - x[1]^2)^2
res <- optim(1:2, r)

tidy(res)
#> # A tibble: 2 x 2
#>   parameter  value
#>   <chr>      <dbl>
#> 1 parameter1  3.00
#> 2 parameter2  9.00
glance(res)
#> # A tibble: 1 x 4
#>          value function.count gradient.count convergence
#>          <dbl>          <int>          <int>       <int>
#> 1 0.0000000358            113             NA           0

^{Created on 2018-11-16 by the reprex package (v0.2.1)}

[billdenney]

Whoops! I had broom.mixed not broom loaded. I have an idea for optim() that I'll bring up in a separate issue.

[mattle24]

@IndrajeetPatil I'm working on glance for glmrob objects, but I'm not familiar with the robustbase package. I'm having trouble finding goodness of fit measures to include. Do you have any suggestions? For lmrob objects, I have included r.squared, sigma, and df.residual, but these fields either seem to not exist or are NULL in glmrob.

[IndrajeetPatil]

@mattle24 Sorry for the late reply. Completely missed this in the notifications.
I can't find that information as well for glmrob objects, so maybe not much that can be done about that.

[mattle24]

All good, thanks for checking!

[HedvigS]

(I would also love a function for tidying hurdle objects. Thanks for a great package regardless :) )

[lirpa939]

(I would also love a function for tidying hurdle objects. Thanks for a great package regardless :) )
I would love to see this too!

[IndrajeetPatil]

Closing following #795 (comment)