what causal discrimination is computing

[rangell]

Currently, for causal discrimination, Themis computes the proportion of equivalence classes where there exists a pair of inputs that differ in output, where an equivalence class is the set of inputs are the same for all the characteristics that are not of interest when computing causal discrimination.

In the future, we might want to incorporate the average percent of pair-wise discrimination among equivalence classes, or some other measure that accounts for all of the outputs in a given equivalence class.