fairlib.src.networks.INLP

fairlib.src.networks.INLP.__init__

fairlib.src.networks.INLP.classifier

fairlib.src.networks.INLP.debias

fairlib.src.networks.INLP.debias.debias_by_specific_directions(directions: List[numpy.ndarray], input_dim: int)

the goal of this function is to perform INLP on a set of user-provided directiosn (instead of learning those directions). :param directions: list of vectors, as numpy arrays. :param input_dim: dimensionality of the vectors.

fairlib.src.networks.INLP.debias.get_debiasing_projection(classifier_class, cls_params: Dict, num_classifiers: int, input_dim: int, is_autoregressive: bool, min_accuracy: float, X_train: numpy.ndarray, Y_train: numpy.ndarray, X_dev: numpy.ndarray, Y_dev: numpy.ndarray, by_class=True, Y_train_main=None, Y_dev_main=None, dropout_rate=0, importance_identifier=[]) → numpy.ndarray
Parameters

  • classifier_class – the sklearn classifier class (SVM/Perceptron etc.)

  • cls_params – a dictionary, containing the params for the sklearn classifier

  • num_classifiers – number of iterations (equivalent to number of dimensions to remove)

  • input_dim – size of input vectors

  • is_autoregressive – whether to train the ith classiifer on the data projected to the nullsapces of w1,…,wi-1

  • min_accuracy – above this threshold, ignore the learned classifier

  • X_train – ndarray, training vectors

  • Y_train – ndarray, training labels (protected attributes)

  • X_dev – ndarray, eval vectors

  • Y_dev – ndarray, eval labels (protected attributes)

  • by_class – if true, at each iteration sample one main-task label, and extract the protected attribute only from vectors from this class

  • T_train_main – ndarray, main-task train labels

  • Y_dev_main – ndarray, main-task eval labels

  • dropout_rate – float, default: 0 (note: not recommended to be used with autoregressive=True)

Returns

P, the debiasing projection; rowspace_projections, the list of all rowspace projection; Ws, the list of all calssifiers.

fairlib.src.networks.INLP.debias.get_projection_to_intersection_of_nullspaces(rowspace_projection_matrices: List[numpy.ndarray], input_dim: int)

Given a list of rowspace projection matrices P_R(w_1), …, P_R(w_n), this function calculates the projection to the intersection of all nullspasces of the matrices w_1, …, w_n. uses the intersection-projection formula of Ben-Israel 2013 http://benisrael.net/BEN-ISRAEL-NOV-30-13.pdf: N(w1)∩ N(w2) ∩ … ∩ N(wn) = N(P_R(w1) + P_R(w2) + … + P_R(wn)) :param rowspace_projection_matrices: List[np.array], a list of rowspace projections :param dim: input dim

fairlib.src.networks.INLP.debias.get_rowspace_projection(W: numpy.ndarray) → numpy.ndarray
Parameters

W – the matrix over its nullspace to project

Returns

the projection matrix over the rowspace