Calculate interaction strength

Compute two-way feature interaction strength based on Friedman's H-statistic.

interaction_strength(pd_2d, pd_1d)

Arguments

pd_2d	Data frame containing the 2D partial dependence as returned by get_pd with var = 'var1_var2'.
pd_1d	List of data frames containing the 1D partial dependence for var1 and var2 as returned by get_pd with var = 'var1' and var = 'var2' respectively.

Value

A numeric value between 0 and 1 indicating interaction strength.

Examples

if (FALSE) {
data('mtpl_be')
features <- setdiff(names(mtpl_be), c('id', 'nclaims', 'expo', 'long', 'lat'))
set.seed(12345)
gbm_fit <- gbm::gbm(as.formula(paste('nclaims ~',
                               paste(features, collapse = ' + '))),
                    distribution = 'poisson',
                    data = mtpl_be,
                    n.trees = 50,
                    interaction.depth = 3,
                    shrinkage = 0.1)
gbm_fun <- function(object, newdata) mean(predict(object, newdata, n.trees = object$n.trees, type = 'response'))
pd_2d <- get_pd(mfit = gbm_fit,
                var = 'ageph_coverage',
                grid = tidyr::expand_grid('ageph' %>% get_grid(data = mtpl_be),
                                          'coverage' %>% get_grid(data = mtpl_be)),
                data = mtpl_be,
                subsample = 10000,
                fun = gbm_fun)
pd_1d <- list(get_pd(mfit = gbm_fit,
                     var = 'ageph',
                     grid = 'ageph' %>% get_grid(data = mtpl_be),
                     data = mtpl_be,
                     subsample = 10000,
                     fun = gbm_fun),
              get_pd(mfit = gbm_fit,
                     var = 'coverage',
                     grid = 'coverage' %>% get_grid(data = mtpl_be),
                     data = mtpl_be,
                     subsample = 10000,
                     fun = gbm_fun))
interaction_strength(pd_2d, pd_1d)
}