Permalink
Fetching contributors…
Cannot retrieve contributors at this time
215 lines (182 sloc) 9.61 KB
# Original work Copyright (c) 2017 Geoff Pleiss
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
#
# Modified work Copyright 2018 IBM Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may not
# use this file except in compliance with the License. You may obtain a copy of
# the License at http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software distributed
# under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
# CONDITIONS OF ANY KIND, either express or implied. See the License for the
# specific language governing permissions and limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import numpy as np
# from scipy.optimize import linprog
from aif360.algorithms import Transformer
from aif360.metrics import ClassificationMetric, utils
class CalibratedEqOddsPostprocessing(Transformer):
"""Calibrated equalized odds postprocessing is a post-processing technique
that optimizes over calibrated classifier score outputs to find
probabilities with which to change output labels with an equalized odds
objective [7]_.
References:
.. [7] G. Pleiss, M. Raghavan, F. Wu, J. Kleinberg, and
K. Q. Weinberger, "On Fairness and Calibration," Conference on Neural
Information Processing Systems, 2017
Adapted from:
https://github.com/gpleiss/equalized_odds_and_calibration/blob/master/calib_eq_odds.py
"""
def __init__(self, unprivileged_groups, privileged_groups,
cost_constraint='weighted', seed=None):
"""
Args:
unprivileged_groups (dict or list(dict)): Representation for
unprivileged group.
privileged_groups (dict or list(dict)): Representation for
privileged group.
cost_contraint: fpr, fnr or weighted
seed (int, optional): Seed to make `predict` repeatable.
"""
super(CalibratedEqOddsPostprocessing, self).__init__(
unprivileged_groups=unprivileged_groups,
privileged_groups=privileged_groups,
seed=seed)
self.seed = seed
self.model_params = None
self.unprivileged_groups = [unprivileged_groups] \
if isinstance(unprivileged_groups, dict) else unprivileged_groups
self.privileged_groups = [privileged_groups] \
if isinstance(privileged_groups, dict) else privileged_groups
self.cost_constraint = cost_constraint
if self.cost_constraint == 'fnr':
self.fn_rate = 1
self.fp_rate = 0
elif self.cost_constraint == 'fpr':
self.fn_rate = 0
self.fp_rate = 1
elif self.cost_constraint == 'weighted':
self.fn_rate = 1
self.fp_rate = 1
self.base_rate_priv = 0.0
self.base_rate_unpriv = 0.0
def fit(self, dataset_true, dataset_pred):
"""Compute parameters for equalizing generalized odds using true and
predicted scores, while preserving calibration.
Args:
dataset_true (BinaryLabelDataset): Dataset containing true `labels`.
dataset_pred (BinaryLabelDataset): Dataset containing predicted
`scores`.
Returns:
CalibratedEqOddsPostprocessing: Returns self.
"""
# Create boolean conditioning vectors for protected groups
cond_vec_priv = utils.compute_boolean_conditioning_vector(
dataset_pred.protected_attributes,
dataset_pred.protected_attribute_names,
self.privileged_groups)
cond_vec_unpriv = utils.compute_boolean_conditioning_vector(
dataset_pred.protected_attributes,
dataset_pred.protected_attribute_names,
self.unprivileged_groups)
cm = ClassificationMetric(dataset_true, dataset_pred,
unprivileged_groups=self.unprivileged_groups,
privileged_groups=self.privileged_groups)
self.base_rate_priv = cm.base_rate(privileged=True)
self.base_rate_unpriv = cm.base_rate(privileged=False)
# Create a dataset with "trivial" predictions
dataset_trivial = dataset_pred.copy(deepcopy =True)
dataset_trivial.scores[cond_vec_priv] = cm.base_rate(privileged=True)
dataset_trivial.scores[cond_vec_unpriv] = cm.base_rate(privileged=False)
cm_triv = ClassificationMetric(dataset_true, dataset_trivial,
unprivileged_groups=self.unprivileged_groups,
privileged_groups=self.privileged_groups)
if self.fn_rate == 0:
priv_cost = cm.generalized_false_positive_rate(privileged=True)
unpriv_cost = cm.generalized_false_positive_rate(privileged=False)
priv_trivial_cost = cm_triv.generalized_false_positive_rate(privileged=True)
unpriv_trivial_cost = cm_triv.generalized_false_positive_rate(privileged=False)
elif self.fp_rate == 0:
priv_cost = cm.generalized_false_negative_rate(privileged=True)
unpriv_cost = cm.generalized_false_negative_rate(privileged=False)
priv_trivial_cost = cm_triv.generalized_false_negative_rate(privileged=True)
unpriv_trivial_cost = cm_triv.generalized_false_negative_rate(privileged=False)
else:
priv_cost = weighted_cost(self.fp_rate, self.fn_rate, cm, privileged=True)
unpriv_cost = weighted_cost(self.fp_rate, self.fn_rate, cm, privileged=False)
priv_trivial_cost = weighted_cost(self.fp_rate, self.fn_rate, cm_triv, privileged=True)
unpriv_trivial_cost = weighted_cost(self.fp_rate, self.fn_rate, cm_triv, privileged=False)
unpriv_costs_more = unpriv_cost > priv_cost
self.priv_mix_rate = (unpriv_cost - priv_cost) / (priv_trivial_cost - priv_cost) if unpriv_costs_more else 0
self.unpriv_mix_rate = 0 if unpriv_costs_more else (priv_cost - unpriv_cost) / (unpriv_trivial_cost - unpriv_cost)
return self
def predict(self, dataset):
"""Perturb the predicted scores to obtain new labels that satisfy
equalized odds constraints, while preserving calibration.
Args:
dataset (BinaryLabelDataset): Dataset containing `scores` that needs
to be transformed.
Returns:
dataset (BinaryLabelDataset): transformed dataset.
"""
if self.seed is not None:
np.random.seed(self.seed)
else:
np.random.set_state(np.random.get_state())
cond_vec_priv = utils.compute_boolean_conditioning_vector(
dataset.protected_attributes,
dataset.protected_attribute_names,
self.privileged_groups)
cond_vec_unpriv = utils.compute_boolean_conditioning_vector(
dataset.protected_attributes,
dataset.protected_attribute_names,
self.unprivileged_groups)
priv_indices = np.random.permutation(sum(cond_vec_priv))[
:int(self.priv_mix_rate * sum(cond_vec_priv))]
priv_new_pred = dataset.scores[cond_vec_priv].copy()
priv_new_pred[priv_indices] = self.base_rate_priv
unpriv_indices = np.random.permutation(sum(cond_vec_unpriv))[
:int(self.unpriv_mix_rate * sum(cond_vec_unpriv))]
unpriv_new_pred = dataset.labels[cond_vec_unpriv].copy()
unpriv_new_pred[unpriv_indices] = self.base_rate_unpriv
dataset_new = dataset.copy(deepcopy=True)
new_scores = np.zeros_like(dataset.scores, dtype=np.float64)
new_scores[cond_vec_priv] = priv_new_pred
new_scores[cond_vec_unpriv] = unpriv_new_pred
dataset_new.scores = new_scores
# Create labels from scores using a default threshold
return dataset_new
def fit_predict(self, dataset_true, dataset_pred):
"""fit and predict methods sequentially."""
return self.fit(dataset_true, dataset_pred).predict(dataset_pred)
######### SUPPORTING FUNCTIONS ##########
def weighted_cost(fp_rate, fn_rate, cm, privileged):
norm_const = float(fp_rate + fn_rate) if\
(fp_rate != 0 and fn_rate != 0) else 1
return ((fp_rate / norm_const
* cm.generalized_false_positive_rate(privileged=privileged)
* (1 - cm.base_rate(privileged=privileged))) + \
(fn_rate / norm_const
* cm.generalized_false_negative_rate(privileged=privileged)
* (1 - cm.base_rate(privileged=privileged))))