inFairness.fairalgo.senstir module

class inFairness.fairalgo.senstir.SenSTIR(network: Module, distance_x: MahalanobisDistances, distance_y: MahalanobisDistances, rho: float, eps: float, auditor_nsteps: int, auditor_lr: float, monte_carlo_samples_ndcg: int)

Bases: Module

Implementes the Sensitive Subspace Robustness (SenSR) algorithm.

Proposed in Individually Fair Ranking

Parameters:

• network (torch.nn.Module) – Network architecture

• distance_x (inFairness.distances.Distance) – Distance metric in the input space

• distance_y (inFairness.distances.Distance) – Distance metric in the output space

• rho (float) – \(\rho\) parameter in the SenSTIR algorithm (see Algorithm 1)

• eps (float) – \(\epsilon\) parameter in the SenSTIR algorithm (see Algorithm 1)

• auditor_nsteps (int) – Number of update steps for the auditor to find worst-case examples

• auditor_lr (float) – Learning rate for the auditor

• monte_carlo_samples_ndcg (int) – Number of monte carlo samples required to estimate the gradient of the empirical version of expectation defined in equation SenSTIR in the reference

forward(Q, relevances, **kwargs)

Defines the computation performed at every call.

Parameters:

• X (torch.Tensor) – Input data

• Y (torch.Tensor) – Expected output data

Returns:

output – Model output

Return type:

torch.Tensor

forward_test(Q)

Forward method during the test phase

forward_train(Q, relevances)

training: bool