DemographicParityDifference#

class ignite.metrics.fairness.DemographicParityDifference(groups, is_multilabel=False, output_transform=<function DemographicParityDifference.<lambda>>, device=device(type='cpu'))[source]#

Calculates the Demographic Parity Difference.

This metric computes the selection rate (the rate of positive predictions) for each unique subgroup in the dataset and returns the maximum difference in selection rates between any two subgroups.

A lower value indicates that the model predicts the positive outcome at roughly the same rate across all subgroups, a standard definition of fairness. This metric is referred to as Group Fairness / Statistical Parity in the fairness literature.

update must receive output of the form (y_pred, y, group_labels) or {'y_pred': y_pred, 'y': y, 'group_labels': group_labels}.
y_pred must be in the following shape (batch_size, num_categories, …) or (batch_size, …).
y must be in the following shape (batch_size, …).
group_labels must be a 1D tensor of shape (batch_size,) containing discrete labels.

Parameters:

groups (Sequence[int]) – a sequence of unique group identifiers.
is_multilabel (bool) – if True, multilabel selection rate is calculated. By default, False.
output_transform (Callable) – a callable that is used to transform the Engine’s process_function’s output into the form expected by the metric.
device (device | str) – specifies which device updates are accumulated on. Setting the metric’s device to be the same as your update arguments ensures the update method is non-blocking. By default, CPU.

Examples

To use with Engine and process_function, simply attach the metric instance to the engine. The output of the engine’s process_function needs to be in the format of (y_pred, y, group_labels).

from collections import OrderedDict

import torch
from torch import nn, optim

from ignite.engine import *
from ignite.handlers import *
from ignite.metrics import *
from ignite.metrics.clustering import *
from ignite.metrics.fairness import *
from ignite.metrics.rec_sys import *
from ignite.metrics.regression import *
from ignite.utils import *

# create default evaluator for doctests

def eval_step(engine, batch):
    return batch

default_evaluator = Engine(eval_step)

# create default optimizer for doctests

param_tensor = torch.zeros([1], requires_grad=True)
default_optimizer = torch.optim.SGD([param_tensor], lr=0.1)

# create default trainer for doctests
# as handlers could be attached to the trainer,
# each test must define his own trainer using `.. testsetup:`

def get_default_trainer():

    def train_step(engine, batch):
        return batch

    return Engine(train_step)

# create default model for doctests

default_model = nn.Sequential(OrderedDict([
    ('base', nn.Linear(4, 2)),
    ('fc', nn.Linear(2, 1))
]))

manual_seed(666)

metric = DemographicParityDifference(groups=[0, 1])
metric.attach(default_evaluator, 'demographic_parity_diff')

# Predictions for 4 items:
# Items 1 and 2 are predicted as class 1 (index 1 has highest prob)
# Items 3 and 4 are predicted as class 0 (index 0 has highest prob)
y_pred = torch.tensor([[0.1, 0.9], [0.2, 0.8], [0.9, 0.1], [0.9, 0.1]])

# Targets (Not actually used for parity, but required by API)
y_true = torch.tensor([1, 1, 0, 0])

# Subgroups
group_labels = torch.tensor([0, 0, 1, 1])

# Subgroup 0: 2 positive predictions / 2 total = 1.0 selection rate
# Subgroup 1: 0 positive predictions / 2 total = 0.0 selection rate

state = default_evaluator.run([[y_pred, y_true, group_labels]])
print(state.metrics['demographic_parity_diff'])

1.0

New in version 0.5.4.

References

Verma & Rubin, Fairness Definitions Explained, 2018.

Methods

DemographicParityDifference#

Search Docs