vi_model for glm should select z value instead of t value

[RoelVerbelen]

The function vip:::vi_model.lm is also used for glm type models, but always selects the "t value" column. The relevant line of code inside this function:

tib <- tibble::tibble(Variable = rownames(coefs), 
    Importance = abs(coefs[, "t value"]), 
    Sign = ifelse(sign(coefs[, "Estimate"]) ==1, yes = "POS", no = "NEG"))

When the family argument is not gaussian, the column name is "z value" instead of "t value", which causes vip:::vi_model.lm to break. Minimal example:

df <- data.frame(norm = rnorm(30), pois = rpois(30, 1), unif = runif(30))

gaussian_fit <- glm(norm ~ unif, family = "gaussian", data = df)
summary(gaussian_fit)$coefficients
#>               Estimate Std. Error    t value  Pr(>|t|)
#> (Intercept)  0.3453485  0.4902475  0.7044371 0.4869788
#> unif        -0.9664243  0.7295934 -1.3246067 0.1960119
vip::vi(gaussian_fit)
#> # A tibble: 1 x 3
#>   Variable Importance Sign 
#>   <chr>         <dbl> <chr>
#> 1 unif           1.32 NEG

poisson_fit <- glm(pois ~ unif, family = "poisson", data = df)
summary(poisson_fit)$coefficients
#>               Estimate Std. Error   z value  Pr(>|z|)
#> (Intercept)  0.6808461  0.3340880  2.037924 0.0415575
#> unif        -0.7225532  0.5352184 -1.350016 0.1770109
vip::vi(poisson_fit)
#> Error in coefs[, "t value"]: subscript out of bounds

^{Created on 2019-10-31 by the reprex package (v0.3.0)}

The logic behind z/t value is in stats::summary.glm. The easiest fix is to select by location instead: the z/t value is always in column 3.