Parameters for regularization learning networks

These parameters are used with regularization learning networks (RLN), such as tabular_rln() when fit with the brulee engine.

Usage

penalty_average(range = c(-15, -5), trans = scales::transform_log10())

step_rate(range = c(0, 8), trans = scales::transform_log10())

penalty_type(values = values_penalty_type)

values_penalty_type

Arguments

range: A two-element vector holding the defaults for the smallest and largest possible values, respectively. If a transformation is specified, these values should be in the transformed units.
trans: A trans object from the scales package, such as scales::transform_log10() or scales::transform_reciprocal(). If not provided, the default is used which matches the units used in range. If no transformation, NULL.
values: A character vector of possible values.

Details

penalty_average(): The target geometric mean of the per-weight regularization coefficients (Theta in Shavitt and Segal (2018)). Best tuned on the log10 scale.
step_rate(): The step size used to update the per-weight regularization coefficients (nu in Shavitt and Segal (2018)). Best tuned on the log10 scale.
penalty_type(): The type of regularization norm applied to per-weight coefficients. L1 is recommended by the original paper.

For penalty_average() and step_rate(), the value is passed to brulee::brulee_rln() on the natural scale.

References

Shavitt, I., & Segal, E. (2018). Regularization learning networks: Deep learning for tabular datasets. Advances in Neural Information Processing Systems, 31, 1379-1389.

Examples

penalty_average()
#> Penalty Average (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [-15, -5]
step_rate()
#> Step Rate (quantitative)
#> Transformer: log-10 [1e-100, Inf]
#> Range (transformed scale): [0, 8]
values_penalty_type
#> [1] "L1" "L2"
penalty_type()
#> Penalty Type (qualitative)
#> 2 possible values include:
#> 'L1' and 'L2'