def bias_grid(Y):
num_stages = Y.columns.size - 1
f, axes = plt.subplots(1, num_stages, figsize=(2+4*num_stages, 12), squeeze=True, sharey=True)
for ax, stage, prev in zip(axes, Y.columns[1:], Y.columns):
rates = Y[Y[prev]][stage].groupby(level=['race', 'gender']).apply(base_rate)
sns.heatmap(rates.unstack(), annot=True, fmt='.1%', cmap='RdBu',
center=base_rate(Y[Y[prev]][stage]), robust=True,
cbar=False, square=True, ax=ax);
ax.set_title(f'{prev} -> {stage}')
plt.close()
return f
def _weighted_cost(self, y_true, probas_pred, pos_label=1,
sample_weight=None):
"""Evaluates the cost function specified by ``self.cost_constraint``."""
fpr = generalized_fpr(y_true, probas_pred, pos_label, sample_weight)
fnr = generalized_fnr(y_true, probas_pred, pos_label, sample_weight)
br = base_rate(y_true, probas_pred, pos_label, sample_weight)
if self.cost_constraint == 'fpr':
return fpr
elif self.cost_constraint == 'fnr':
return fnr
elif self.cost_constraint == 'weighted':
return fpr * (1 - br) + fnr * br
else:
raise ValueError("`cost_constraint` must be one of: 'fpr', 'fnr', "
"or 'weighted'")
def fit(self, X, y, labels=None, pos_label=1, sample_weight=None):
"""Compute the mixing rates required to satisfy the cost constraint.
Args:
X (array-like): Probability estimates of the targets as returned by
a ``predict_proba()`` call or equivalent.
y (pandas.Series): Ground-truth (correct) target values.
labels (list, optional): The ordered set of labels values. Must
match the order of columns in X if provided. By default,
all labels in y are used in sorted order.
pos_label (scalar, optional): The label of the positive class.
sample_weight (array-like, optional): Sample weights.
Returns:
self
"""
X, y, sample_weight = check_inputs(X, y, sample_weight)
groups, self.prot_attr_ = check_groups(y, self.prot_attr,
ensure_binary=True)
self.classes_ = labels if labels is not None else np.unique(y)
self.groups_ = np.unique(groups)
self.pos_label_ = pos_label
if len(self.classes_) > 2:
raise ValueError('Only binary classification is supported.')
if pos_label not in self.classes_:
raise ValueError('pos_label={} is not in the set of labels. The '
'valid values are:\n{}'.format(pos_label, self.classes_))
X = X[:, np.nonzero(self.classes_ == self.pos_label_)[0][0]]
# local function to return corresponding args for metric evaluation
def _args(grp_idx, triv=False):
idx = (groups == self.groups_[grp_idx])
pred = np.full_like(X, self.base_rates_[grp_idx]) if triv else X
return [y[idx], pred[idx], pos_label, sample_weight[idx]]
self.base_rates_ = [base_rate(*_args(i)) for i in range(2)]
costs = [self._weighted_cost(*_args(i)) for i in range(2)]
self.mix_rates_ = [(costs[1] - costs[0])
/ (self._weighted_cost(*_args(0, True)) - costs[0]),
(costs[0] - costs[1])
/ (self._weighted_cost(*_args(1, True)) - costs[1])]
self.mix_rates_[np.argmax(costs)] = 0
return self
def test_base_rate():
"""Tests that the old and new base_rate matches exactly."""
base = base_rate(y, y_pred, sample_weight=sample_weight)
assert base == cm.base_rate()