man/instructions/adding_new_metric.md

In fairmodels: Flexible Tool for Bias Detection, Visualization, and Mitigation

Instruction for adding one custom metric to fairmodels.

1. Make a fork of the repository
2. In file R/calculate_group_fairness_metrics.R change number of columns to 13 in "group_metric_matrix". Add name of metric to "rownames(group_metric_matrix)". Add your metric by defining it with fields from confusion matrix (tp, tn, fp, fn). Add it to "group_metric_matrix".
3. In file R/fairness_check.R in place where "parity_loss_metric_data" is defined, change number of columns to 13.
4. In file R/heper_functions.R add your metrics name to "parity_loss_metrics"

That is all. Now when creating fairness object your new metric will be visible in fileds parity_loss_metric_data and groups_data. To visualize your metric, just add its name to parameter responsible for fairness metric, like in here: stack_metrics(fobject, fairness_metrics = c("ACC", "TPR", "PPV", "FPR", "STP", "NEW_METRIC")) %>% plot()

fairmodels documentation built on Aug. 24, 2022, 1:05 a.m.