In parsnip: A Common API to Modeling and Analysis Functions

r descr_models("ordinal_reg", "ordinalNet")

Tuning Parameters

defaults <- 
  tibble::tibble(
    parsnip = c("ordinal_link", "odds_link", "penalty", "mixture"),
    default = c("logit", "cumulative", "see below", "1.0")
  )

param <-
  ordinal_reg() |>
  set_engine("ordinalNet") |>
  make_parameter_list(defaults)

This model has r nrow(param) tuning parameters:

param$item

Translation from parsnip to the original package

r uses_extension("ordinal_reg", "ordinalNet", "classification")

library(ordered)

ordinal_reg(
  penalty = double(0),
  mixture = double(0),
  ordinal_link = character(0),
  odds_link = character(0)
) %>%
  set_engine("ordinalNet") %>%
  # "classification" is the only mode
  translate()

Controlling penalty values

ordinalNet(), like glmnet(), simultaneously computes a set of parameter estimates for multiple penalty values. Predictions can be made at these penalty values at the same time. However, unlike glmnet(), ordinalNet() does not interpolate if you want to predict using penalty values not exactly among those it precomputed. Similarly, it cannot predict for models with penalties outside of the range of those precomputed.

The \pkg{ordered} package can interpolate between the preset penalty values but cannot predict outside of their range; this will cause an error.

We suggest that you set the collection of penalty values when fitting the model. This is important when tuning the model. Different data sets and mixture values (a.k.a. alpha) will pair best with different sets of penalties and it might be good to set a wide range.

To do this, you can use set_engine() to pass a vector of penalty values as so:

# Example of setting a wide penalty range
penalties <- 10^seq(-10, 0, length.out = 20)

ordinal_reg(penalty = tune()) |>
  set_engine("ordinalNet", path_values = !!penalties)

See [glmnet-details] for more background.

Preprocessing requirements

#| child: template-makes-dummies.Rmd

#| child: template-same-scale.Rmd

By default, [ordinalNet::ordinalNet()] uses the argument standardize = TRUE to center and scale the data.

Case weights

#| child: template-uses-case-weights.Rmd

Prediction types

#| label: predict-types
parsnip:::get_from_env("ordinal_reg_predict") |>
  dplyr::filter(engine == "ordinalNet") |>
  dplyr::select(mode, type)

References

Wurm MJ, Rathouz PJ, Hanlon BM. 2021. Regularized Ordinal Regression and the ordinalNet R Package. Journal of Statistical Software 99(6):1-42. \doi{10.18637/jss.v099.i06}.