Fairness Report

In mlr3fairness: Fairness Auditing and Debiasing for 'mlr3'

knitr::opts_chunk$set(warning = FALSE, message = FALSE, echo=FALSE)
library("mlr3")
library("mlr3viz")
library("mlr3fairness")
library("ggplot2")
library("kableExtra")

Audit Report: Summary

This document introduces on how to use mlr3fairness to create audit reports with different tasks throughout the fairness exploration process.

There are three main sections for this document. Which describe the details for the task, the model and the interpretability of the parameters.

Jump to section:

• Task details
• Model details
• Interpretability

Task details

In this fairness report, we investigate the fairness of the following task:

task

Task Documentation:

Here we the basic details for the task.

task_summary(task) %>%
  kbl() %>%
  kable_paper("hover", full_width = F)

Exploratory Data Analysis:

We also report the number of missing values, types and the levels for each feature:

df_summary = data.frame(task$col_info)
df_summary = df_summary[df_summary$id %in% c(task$feature_names, task$target_names), ]
df_summary %>%
  cbind(data.frame("Missings (%)" = task$missings() / task$nrow)) %>%
  kbl() %>%
  kable_paper("hover", full_width = F)

We first look at the label distribution:

autoplot(task) + facet_wrap(task$col_roles$pta)

Model details

We could see the model that has been used in resample_result:

resample_result$learner

Fairness Metrics

We furthermore report more than one fairness metric. Below metrics are the mean across all the resample results.

fair_metrics = msrs(c("fairness.acc","fairness.eod","fairness.fnr", "fairness.fpr","fairness.ppv", "fairness.cv"))

resample_result$aggregate(fair_metrics) %>%
  kbl(col.names = c("value")) %>%
  kable_paper("hover", full_width = F)

We can furthermore employ several visualizations to report the fairness. For example, the fairness and accuracy trade off, compare metrics visualization and the fairness prediction density of our model. For more detailed usage and examples, you may want to check the visualization vignette.

fairness_accuracy_tradeoff(resample_result, msr("fairness.fnr"))

fairness_prediction_density(resample_result)

compare_metrics(resample_result, fair_metrics)

Interpretability

Finally, we use the external package to gain further insight into our model. For the following example we use the iml package as a demonstration. We need first extract the learner from resample_result and wrap it in a Predictor object.

You could generate the variable importance plot like this

library("iml")

target = task$target_names
twocols = task$feature_names[1:2]
learner = resample_result$learner
learner$train(task)

model = Predictor$new(model = learner,
                      data = task$data()[,.SD, .SDcols = !target],
                      y = task$data()[, ..target])

imp <- FeatureImp$new(model, loss = "ce")
plot(imp)

Or generate the feature effects plot:

effect = FeatureEffects$new(model, method = "pdp", grid.size = 10)
effect$plot(features = twocols)

For more details on interpretability, check the documentation of the iml package.

mlr3fairness documentation built on May 31, 2023, 7:22 p.m.