Counterfactual-pair debiasing and the gender direction (Eq 6.18-6.19)

Usage

morie_kamath_ch6_debias_regularizer(A, E, lam)

morie_kamath_ch6_gender_direction(A, E)

Arguments

A: List of two-element word pairs.
E: Embedding map/function.
lam: Non-negative weight.

Value

List with estimate, per_pair.

Details

One pair reader serves both Python modules exactly as km095 imports km094's _pair_vectors: km094 -> morie_kamath_ch6_debias_regularizer (lam * sum of squared within-pair distances), km095 -> morie_kamath_ch6_gender_direction (mean of E(second) - E(first) over the pairs).

Examples

E <- list(m1 = c(1, 0), f1 = c(0, 1), m2 = c(2, 0), f2 = c(0,
    2))
P <- list(c("m1", "f1"), c("m2", "f2"))
morie_kamath_ch6_debias_regularizer(P, E, 0.5)
#> $estimate
#> [1] 5
#> 
#> $per_pair
#> [1] 2 8
#> 
#> $unweighted
#> [1] 10
#> 
#> $lam
#> [1] 0.5
#> 
#> $n
#> [1] 2
#> 
#> $method
#> [1] "counterfactual-pair debiasing regulariser (Kamath Eq 6.18)"
#>