calc_balanced_precision: Calculate balanced precision given actual and baseline...

In SchlossLab/mikropml: User-Friendly R Package for Supervised Machine Learning Pipelines

Calculate balanced precision given actual and baseline precision

Description

Implements Equation 1 from Wu et al. 2021 \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/j.ajhg.2021.08.012")}. It is the same as Equation 7 if AUPRC (aka prAUC) is used in place of precision.

Usage

calc_balanced_precision(precision, prior)

Arguments

precision	actual precision of the model.
prior	baseline precision, aka frequency of positives. Can be calculated with calc_baseline_precision

Value

the expected precision if the data were balanced

Author(s)

Kelly Sovacool [e-mail]

Examples

prior <- calc_baseline_precision(otu_mini_bin,
  outcome_colname = "dx",
  pos_outcome = "cancer"
)
calc_balanced_precision(otu_mini_bin_results_rf$performance$Precision, prior)

otu_mini_bin_results_rf$performance %>%
  dplyr::mutate(
    balanced_precision = calc_balanced_precision(Precision, prior),
    aubprc = calc_balanced_precision(prAUC, prior)
  ) %>%
  dplyr::select(AUC, Precision, balanced_precision, aubprc)

# cumulative performance for a single model
sensspec_1 <- calc_model_sensspec(
  otu_mini_bin_results_glmnet$trained_model,
  otu_mini_bin_results_glmnet$test_data,
  "dx"
)
head(sensspec_1)
prior <- calc_baseline_precision(otu_mini_bin,
  outcome_colname = "dx",
  pos_outcome = "cancer"
)
sensspec_1 %>%
  dplyr::mutate(balanced_precision = calc_balanced_precision(precision, prior)) %>%
  dplyr::rename(recall = sensitivity) %>%
  calc_mean_perf(group_var = recall, sum_var = balanced_precision) %>%
  plot_mean_prc(ycol = mean_balanced_precision)

SchlossLab/mikropml documentation built on Aug. 24, 2023, 9:51 p.m.