apunkt/ai-privacy-toolkit
1
0
Fork 0
mirror of https://github.com/IBM/ai-privacy-toolkit.git synced 2026-04-25 04:46:21 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix issue with computed ranges for one-hot encoded features (#90)

Browse source 
Signed-off-by: abigailt <abigailt@il.ibm.com>
This commit is contained in:
abigailgold 2024-01-17 12:45:22 -05:00 • committed by GitHub
parent d8de062d43
commit a8f5326572
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 46 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
apt/minimization/minimizer.py
View file

@ -576,7 +576,7 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
if feature not in feature_data.keys():
fd = {}
values = list(x.loc[:, feature])
if feature not in self.categorical_features:
if feature not in self.categorical_features and feature not in self.all_one_hot_features:
fd['min'] = min(values)
fd['max'] = max(values)
fd['range'] = max(values) - min(values)

45
tests/test_minimizer.py
View file

@ -965,6 +965,51 @@ def test_minimizer_ndarray_one_hot():
assert ((np.min(transformed_slice, axis=1) == 0).all())
def test_minimizer_ndarray_one_hot_single_value():
x_train = np.array([[23, 0, 1, 0, 165],
[45, 0, 1, 0, 158],
[56, 1, 0, 0, 123],
[67, 0, 1, 0, 154],
[45, 1, 0, 0, 149],
[42, 1, 0, 0, 166],
[73, 0, 1, 0, 172],
[94, 0, 1, 0, 168],
[69, 0, 1, 0, 175],
[24, 1, 0, 0, 181],
[18, 1, 0, 0, 190]])
y_train = np.array([1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 0])
model = DecisionTreeClassifier()
model.fit(x_train, y_train)
predictions = model.predict(x_train)
features = ['0', '1', '2', '3', '4']
QI = [0, 1, 2, 3]
QI_slices = [[1, 2, 3]]
target_accuracy = 0.7
gen = GeneralizeToRepresentative(model, target_accuracy=target_accuracy, feature_slices=QI_slices,
features_to_minimize=QI)
gen.fit(dataset=ArrayDataset(x_train, predictions))
transformed = gen.transform(dataset=ArrayDataset(x_train))
gener = gen.generalizations
expected_generalizations = {'categories': {}, 'category_representatives': {},
'range_representatives': {'0': [34.5]}, 'ranges': {'0': [34.5]},
'untouched': ['4', '1', '2', '3']}
compare_generalizations(gener, expected_generalizations)
check_features(features, expected_generalizations, transformed, x_train)
ncp = gen.ncp.transform_score
check_ncp(ncp, expected_generalizations)
rel_accuracy = model.score(transformed, predictions)
assert ((rel_accuracy >= target_accuracy) or (target_accuracy - rel_accuracy) <= ACCURACY_DIFF)
transformed_slice = transformed[:, QI_slices[0]]
assert ((np.sum(transformed_slice, axis=1) == 1).all())
assert ((np.max(transformed_slice, axis=1) == 1).all())
assert ((np.min(transformed_slice, axis=1) == 0).all())
def test_minimizer_ndarray_one_hot_gen():
x_train = np.array([[23, 0, 1, 165],
[45, 0, 1, 158],