mirror of
https://github.com/IBM/ai-privacy-toolkit.git
synced 2026-05-03 00:32:37 +02:00
Train just on qi (#15)
* QI updates * update code to support training ML on QI features * fix code so features that are not from QI should not be part of generalizations and add description * merging two branches, training on QI and on all data * adding tests and asserts
This commit is contained in:
olasaadi
GitHub
parent
2eb626c00c
commit
a9a93c8a3a
4 changed files with 373 additions and 135 deletions
|
|
@ -19,9 +19,8 @@ class Anonymize:
|
|||
"""
|
||||
:param k: The privacy parameter that determines the number of records that will be indistinguishable from each
|
||||
other (when looking at the quasi identifiers). Should be at least 2.
|
||||
:param quasi_identifiers: The indexes of the features that need to be anonymized (these should be the features
|
||||
that may directly, indirectly or in combination with additional data, identify an
|
||||
individual).
|
||||
:param quasi_identifiers: The features that need to be minimized in case of pandas data, and indexes of features
|
||||
in case of numpy data.
|
||||
:param categorical_features: The list of categorical features (should only be supplied when passing data as a
|
||||
pandas dataframe.
|
||||
"""
|
||||
|
|
|
|||
|
|
@ -47,17 +47,22 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
|
||||
features : list of str, optional
|
||||
The feature names, in the order that they appear in the data.
|
||||
|
||||
categorical_features: list of str, optional
|
||||
The list of categorical features should only be supplied when
|
||||
passing data as a pandas dataframe.
|
||||
|
||||
features_to_minimize: List of str or numbers, optional
|
||||
The features that need to be minimized in case of pandas data,
|
||||
and indexes of features in case of numpy data.
|
||||
cells : list of object, optional
|
||||
The cells used to generalize records. Each cell must define a
|
||||
range or subset of categories for each feature, as well as a
|
||||
representative value for each feature.
|
||||
This parameter should be used when instantiating a transformer
|
||||
object without first fitting it.
|
||||
train_only_QI : Bool, optional
|
||||
The required method to train data set for minimizing. Default is
|
||||
to train the tree just on the features that are given as
|
||||
features_to_minimize.
|
||||
|
||||
Attributes
|
||||
----------
|
||||
|
|
@ -78,7 +83,8 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
"""
|
||||
|
||||
def __init__(self, estimator=None, target_accuracy=0.998, features=None,
|
||||
cells=None, categorical_features=None):
|
||||
cells=None, categorical_features=None, features_to_minimize: Union[np.ndarray, list] = None
|
||||
, train_only_QI=True):
|
||||
self.estimator = estimator
|
||||
self.target_accuracy = target_accuracy
|
||||
self.features = features
|
||||
|
|
@ -86,8 +92,9 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
self.categorical_features = []
|
||||
if categorical_features:
|
||||
self.categorical_features = categorical_features
|
||||
self.is_numpy = True
|
||||
|
||||
self.features_to_minimize = features_to_minimize
|
||||
self.train_only_QI = train_only_QI
|
||||
|
||||
def get_params(self, deep=True):
|
||||
"""Get parameters for this estimator.
|
||||
|
||||
|
|
@ -201,17 +208,33 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
self.cells_ = {}
|
||||
self.categorical_values = {}
|
||||
|
||||
if self.is_numpy:
|
||||
X = pd.DataFrame(X, columns=self._features)
|
||||
|
||||
|
||||
# Going to fit
|
||||
# (currently not dealing with option to fit with only X and y and no estimator)
|
||||
if self.estimator and X is not None and y is not None:
|
||||
|
||||
if self.is_numpy:
|
||||
if not self.features_to_minimize:
|
||||
self.features_to_minimize = [i for i in range(len(self._features))]
|
||||
x_QI = X[:, self.features_to_minimize]
|
||||
self.features_to_minimize = [self._features[i] for i in self.features_to_minimize]
|
||||
X = pd.DataFrame(X, columns=self._features)
|
||||
else:
|
||||
if not self.features_to_minimize:
|
||||
self.features_to_minimize = self._features
|
||||
x_QI = X.loc[:, self.features_to_minimize]
|
||||
x_QI = pd.DataFrame(x_QI, columns=self.features_to_minimize)
|
||||
# divide dataset into train and test
|
||||
used_data = X
|
||||
if self.train_only_QI:
|
||||
used_data = x_QI
|
||||
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y,
|
||||
test_size=0.4,
|
||||
random_state=18)
|
||||
X_train_QI = X_train.loc[:, self.features_to_minimize]
|
||||
X_test_QI = X_test.loc[:, self.features_to_minimize]
|
||||
used_X_train = X_train
|
||||
if self.train_only_QI:
|
||||
used_X_train = X_train_QI
|
||||
|
||||
# collect feature data (such as min, max)
|
||||
|
||||
|
|
@ -229,42 +252,71 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
feature_data[feature] = fd
|
||||
|
||||
# prepare data for DT
|
||||
categorical_features = list(self.categorical_features)
|
||||
categorical_features = [f for f in self._features if f in self.categorical_features and
|
||||
f in self.features_to_minimize]
|
||||
|
||||
|
||||
numeric_transformer = Pipeline(
|
||||
steps=[('imputer', SimpleImputer(strategy='constant', fill_value=0))]
|
||||
)
|
||||
|
||||
# numeric_features = list(self._features) - list(self.categorical_features)
|
||||
numeric_features = [item for item in self._features if item not in self.categorical_features]
|
||||
numeric_features = [f for f in self._features if f not in self.categorical_features and
|
||||
f in self.features_to_minimize]
|
||||
categorical_transformer = OneHotEncoder(handle_unknown="ignore", sparse=False)
|
||||
|
||||
preprocessor = ColumnTransformer(
|
||||
preprocessor_QI_features = ColumnTransformer(
|
||||
transformers=[
|
||||
("num", numeric_transformer, numeric_features),
|
||||
("cat", categorical_transformer, categorical_features),
|
||||
]
|
||||
)
|
||||
preprocessor.fit(X)
|
||||
preprocessor_QI_features.fit(x_QI)
|
||||
|
||||
# preprocessor to fit data that have features not included in QI (to get accuracy)
|
||||
numeric_features = [f for f in self._features if f not in self.categorical_features]
|
||||
numeric_transformer = Pipeline(
|
||||
steps=[('imputer', SimpleImputer(strategy='constant', fill_value=0))]
|
||||
)
|
||||
categorical_transformer = OneHotEncoder(handle_unknown="ignore", sparse=False)
|
||||
preprocessor = ColumnTransformer(
|
||||
transformers=[
|
||||
("num", numeric_transformer, numeric_features),
|
||||
("cat", categorical_transformer, self.categorical_features),
|
||||
]
|
||||
)
|
||||
preprocessor.fit(X)
|
||||
x_prepared = preprocessor.transform(X_train)
|
||||
self.preprocessor = preprocessor
|
||||
if self.train_only_QI:
|
||||
x_prepared = preprocessor_QI_features.transform(X_train_QI)
|
||||
|
||||
self._preprocessor = preprocessor
|
||||
|
||||
self.cells_ = {}
|
||||
self.dt_ = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
self.dt_.fit(x_prepared, y_train)
|
||||
self._modify_categorical_features(X)
|
||||
self._modify_categorical_features(used_data)
|
||||
|
||||
x_prepared = pd.DataFrame(x_prepared, columns=self.categorical_data.columns)
|
||||
|
||||
self._calculate_cells()
|
||||
self._modify_cells()
|
||||
# features that are not from QI should not be part of generalizations
|
||||
for feature in self._features:
|
||||
if feature not in self.features_to_minimize:
|
||||
self._remove_feature_from_cells(self.cells_, self.cells_by_id_, feature)
|
||||
|
||||
nodes = self._get_nodes_level(0)
|
||||
self._attach_cells_representatives(x_prepared, X_train, y_train, nodes)
|
||||
self._attach_cells_representatives(x_prepared, used_X_train, y_train, nodes)
|
||||
|
||||
# self.cells_ currently holds the generalization created from the tree leaves
|
||||
self._calculate_generalizations()
|
||||
|
||||
# apply generalizations to test data
|
||||
x_prepared_test = preprocessor.transform(X_test)
|
||||
if self.train_only_QI:
|
||||
x_prepared_test = preprocessor_QI_features.transform(X_test_QI)
|
||||
|
||||
x_prepared_test = pd.DataFrame(x_prepared_test, index=X_test.index, columns=self.categorical_data.columns)
|
||||
|
||||
generalized = self._generalize(X_test, x_prepared_test, nodes, self.cells_, self.cells_by_id_)
|
||||
|
|
@ -285,8 +337,8 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
cells_by_id_prev = self.cells_by_id_
|
||||
nodes = self._get_nodes_level(level)
|
||||
self._calculate_level_cells(level)
|
||||
self._attach_cells_representatives(x_prepared, used_X_train, y_train, nodes)
|
||||
|
||||
self._attach_cells_representatives(x_prepared, X_train, y_train, nodes)
|
||||
self._calculate_generalizations()
|
||||
generalized = self._generalize(X_test, x_prepared_test, nodes, self.cells_,
|
||||
self.cells_by_id_)
|
||||
|
|
@ -335,7 +387,6 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
X : {array-like, sparse-matrix}, shape (n_samples, n_features), If provided as a pandas dataframe,
|
||||
may contain both numeric and categorical data.
|
||||
The input samples.
|
||||
|
||||
Returns
|
||||
-------
|
||||
X_transformed : numpy or pandas according to the input type, shape (n_samples, n_features)
|
||||
|
|
@ -357,7 +408,6 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
else:
|
||||
self.is_numpy = False
|
||||
|
||||
|
||||
if X.shape[1] != self.n_features_ and self.n_features_ != 0:
|
||||
raise ValueError('Shape of input is different from what was seen'
|
||||
'in `fit`')
|
||||
|
|
@ -428,19 +478,24 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
self.categorical_values = {}
|
||||
self.oneHotVectorFeaturesToFeatures = {}
|
||||
features_to_remove = []
|
||||
used_features = self._features
|
||||
if self.train_only_QI:
|
||||
used_features = self.features_to_minimize
|
||||
for feature in self.categorical_features:
|
||||
try:
|
||||
all_values = X.loc[:, feature]
|
||||
values = list(all_values.unique())
|
||||
self.categorical_values[feature] = values
|
||||
X[feature] = pd.Categorical(X.loc[:, feature], categories=values, ordered=False)
|
||||
ohe = pd.get_dummies(X[feature], prefix=feature)
|
||||
for oneHotVectorFeature in ohe.columns:
|
||||
self.oneHotVectorFeaturesToFeatures[oneHotVectorFeature] = feature
|
||||
X = pd.concat([X, ohe], axis=1)
|
||||
features_to_remove.append(feature)
|
||||
except KeyError:
|
||||
print("feature " + feature + "not found in training data")
|
||||
if feature in used_features:
|
||||
try:
|
||||
all_values = X.loc[:, feature]
|
||||
values = list(all_values.unique())
|
||||
self.categorical_values[feature] = values
|
||||
X[feature] = pd.Categorical(X.loc[:, feature], categories=values, ordered=False)
|
||||
ohe = pd.get_dummies(X[feature], prefix=feature)
|
||||
for oneHotVectorFeature in ohe.columns:
|
||||
self.oneHotVectorFeaturesToFeatures[oneHotVectorFeature] = feature
|
||||
X = pd.concat([X, ohe], axis=1)
|
||||
features_to_remove.append(feature)
|
||||
except KeyError:
|
||||
print("feature " + feature + "not found in training data")
|
||||
|
||||
self.categorical_data = X.drop(features_to_remove, axis=1)
|
||||
|
||||
def _cell_contains_numeric(self, f, range, x):
|
||||
|
|
@ -556,7 +611,7 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
right_child = self.dt_.tree_.children_right[node]
|
||||
left_cell = self.cells_by_id_[left_child]
|
||||
right_cell = self.cells_by_id_[right_child]
|
||||
new_cell = {'id': int(node), 'ranges': {}, 'categories': {},
|
||||
new_cell = {'id': int(node), 'ranges': {}, 'categories': {}, 'untouched': [],
|
||||
'label': None, 'representative': None}
|
||||
for feature in left_cell['ranges'].keys():
|
||||
new_cell['ranges'][feature] = {}
|
||||
|
|
@ -566,6 +621,9 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
new_cell['categories'][feature] = \
|
||||
list(set(left_cell['categories'][feature]) |
|
||||
set(right_cell['categories'][feature]))
|
||||
for feature in left_cell['untouched']:
|
||||
if feature in right_cell['untouched']:
|
||||
new_cell['untouched'].append(feature)
|
||||
self._calculate_level_cell_label(left_cell, right_cell, new_cell)
|
||||
new_cells.append(new_cell)
|
||||
new_cells_by_id[new_cell['id']] = new_cell
|
||||
|
|
@ -675,7 +733,6 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
else:
|
||||
replace = representatives.loc[i].to_frame().T.reset_index(drop=True)
|
||||
replace.index = indexes
|
||||
# replace = self.preprocessor.transform(replace)
|
||||
replace = pd.DataFrame(replace, indexes, columns=self._features)
|
||||
original_data_generalized.loc[indexes, representatives.columns.tolist()] = replace
|
||||
|
||||
|
|
@ -701,8 +758,6 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
if feature is None:
|
||||
return None
|
||||
GeneralizeToRepresentative._remove_feature_from_cells(self.cells_, self.cells_by_id_, feature)
|
||||
# del self.generalizations_['ranges'][feature]
|
||||
# self.generalizations_['untouched'].append(feature)
|
||||
return feature
|
||||
|
||||
def _get_feature_to_remove(self, original_data, prepared_data, nodes, labels, feature_data, current_accuracy):
|
||||
|
|
@ -730,7 +785,7 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
cells_by_id = copy.deepcopy(self.cells_by_id_)
|
||||
GeneralizeToRepresentative._remove_feature_from_cells(new_cells, cells_by_id, feature)
|
||||
generalized = self._generalize(original_data, prepared_data, nodes, new_cells, cells_by_id)
|
||||
accuracy_gain = self.estimator.score(self.preprocessor.transform(generalized),
|
||||
accuracy_gain = self.estimator.score(self._preprocessor.transform(generalized),
|
||||
labels) - current_accuracy
|
||||
if accuracy_gain < 0:
|
||||
accuracy_gain = 0
|
||||
|
|
@ -753,7 +808,7 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
cells_by_id = copy.deepcopy(self.cells_by_id_)
|
||||
GeneralizeToRepresentative._remove_feature_from_cells(new_cells, cells_by_id, feature)
|
||||
generalized = self._generalize(original_data, prepared_data, nodes, new_cells, cells_by_id)
|
||||
accuracy_gain = self.estimator.score(self.preprocessor.transform(generalized),
|
||||
accuracy_gain = self.estimator.score(self._preprocessor.transform(generalized),
|
||||
labels) - current_accuracy
|
||||
|
||||
if accuracy_gain < 0:
|
||||
|
|
@ -923,7 +978,7 @@ class GeneralizeToRepresentative(BaseEstimator, MetaEstimatorMixin, TransformerM
|
|||
cell['untouched'] = []
|
||||
if feature in cell['ranges'].keys():
|
||||
del cell['ranges'][feature]
|
||||
else:
|
||||
elif feature in cell['categories'].keys():
|
||||
del cell['categories'][feature]
|
||||
cell['untouched'].append(feature)
|
||||
cells_by_id[cell['id']] = cell.copy()
|
||||
|
|
|
|||
77
apt/utils.py
77
apt/utils.py
|
|
@ -6,15 +6,15 @@ from os import path, mkdir
|
|||
from six.moves.urllib.request import urlretrieve
|
||||
|
||||
|
||||
def _load_iris(test_set_size: float=0.3):
|
||||
def _load_iris(test_set_size: float = 0.3):
|
||||
iris = datasets.load_iris()
|
||||
data = iris.data
|
||||
labels = iris.target
|
||||
|
||||
# Split training and test sets
|
||||
x_train, x_test, y_train, y_test = model_selection.train_test_split(data, labels, test_size=test_set_size,
|
||||
random_state=18, stratify=labels,
|
||||
shuffle=True)
|
||||
random_state=18, stratify=labels,
|
||||
shuffle=True)
|
||||
|
||||
return (x_train, y_train), (x_test, y_test)
|
||||
|
||||
|
|
@ -29,6 +29,77 @@ def get_iris_dataset():
|
|||
return _load_iris()
|
||||
|
||||
|
||||
def get_german_credit_dataset(test_set: float = 0.3):
|
||||
"""
|
||||
Loads the UCI German_credit dataset from `tests/datasets/german` or downloads it if necessary.
|
||||
|
||||
:return: Dataset and labels as pandas dataframes.
|
||||
"""
|
||||
|
||||
url = 'https://archive.ics.uci.edu/ml/machine-learning-databases/statlog/german/german.data'
|
||||
data_dir = '../datasets/german'
|
||||
data_file = '../datasets/german/data'
|
||||
|
||||
if not path.exists(data_dir):
|
||||
mkdir(data_dir)
|
||||
|
||||
ssl._create_default_https_context = ssl._create_unverified_context
|
||||
if not path.exists(data_file):
|
||||
urlretrieve(url, data_file)
|
||||
|
||||
# load data
|
||||
features = ["Existing_checking_account", "Duration_in_month", "Credit_history", "Purpose", "Credit_amount",
|
||||
"Savings_account", "Present_employment_since", "Installment_rate", "Personal_status_sex", "debtors",
|
||||
"Present_residence", "Property", "Age", "Other_installment_plans", "Housing",
|
||||
"Number_of_existing_credits", "Job", "N_people_being_liable_provide_maintenance", "Telephone",
|
||||
"Foreign_worker", "label"]
|
||||
data = pd.read_csv(data_file, sep=" ", names=features, engine="python")
|
||||
# remove rows with missing label
|
||||
data = data.dropna(subset=["label"])
|
||||
_modify_german_dataset(data)
|
||||
|
||||
# Split training and test sets
|
||||
stratified = sklearn.model_selection.StratifiedShuffleSplit(n_splits=1, test_size=test_set, random_state=18)
|
||||
for train_set, test_set in stratified.split(data, data["label"]):
|
||||
train = data.iloc[train_set]
|
||||
test = data.iloc[test_set]
|
||||
x_train = train.drop(["label"], axis=1)
|
||||
y_train = train.loc[:, "label"]
|
||||
x_test = test.drop(["label"], axis=1)
|
||||
y_test = test.loc[:, "label"]
|
||||
|
||||
categorical_features = ["Existing_checking_account", "Credit_history", "Purpose", "Savings_account",
|
||||
"Present_employment_since", "Personal_status_sex", "debtors", "Property",
|
||||
"Other_installment_plans", "Housing", "Job"]
|
||||
x_train.reset_index(drop=True, inplace=True)
|
||||
y_train.reset_index(drop=True, inplace=True)
|
||||
x_test.reset_index(drop=True, inplace=True)
|
||||
y_test.reset_index(drop=True, inplace=True)
|
||||
|
||||
return (x_train, y_train), (x_test, y_test)
|
||||
|
||||
|
||||
def _modify_german_dataset(data):
|
||||
|
||||
def modify_Foreign_worker(value):
|
||||
if value == 'A201':
|
||||
return 1
|
||||
elif value == 'A202':
|
||||
return 0
|
||||
else:
|
||||
raise Exception('Bad value')
|
||||
|
||||
def modify_Telephone(value):
|
||||
if value == 'A191':
|
||||
return 0
|
||||
elif value == 'A192':
|
||||
return 1
|
||||
else:
|
||||
raise Exception('Bad value')
|
||||
data['Foreign_worker'] = data['Foreign_worker'].apply(modify_Foreign_worker)
|
||||
data['Telephone'] = data['Telephone'].apply(modify_Telephone)
|
||||
|
||||
|
||||
def get_adult_dataset():
|
||||
"""
|
||||
Loads the UCI Adult dataset from `tests/datasets/adult` or downloads it if necessary.
|
||||
|
|
|
|||
|
|
@ -12,8 +12,7 @@ from sklearn.preprocessing import OneHotEncoder, StandardScaler
|
|||
|
||||
from apt.minimization import GeneralizeToRepresentative
|
||||
from sklearn.tree import DecisionTreeClassifier
|
||||
from apt.utils import get_iris_dataset, get_adult_dataset, get_nursery_dataset
|
||||
|
||||
from apt.utils import get_iris_dataset, get_adult_dataset, get_nursery_dataset, get_german_credit_dataset
|
||||
|
||||
@pytest.fixture
|
||||
def data():
|
||||
|
|
@ -43,11 +42,7 @@ def test_minimizer_params(data):
|
|||
gen = GeneralizeToRepresentative(base_est, features=features, cells=cells)
|
||||
gen.fit()
|
||||
transformed = gen.transform(X)
|
||||
expected_transformed = np.array([[26, 149],
|
||||
[58, 163],
|
||||
[31, 184]])
|
||||
assert(np.array_equal(expected_transformed, transformed))
|
||||
|
||||
|
||||
|
||||
def test_minimizer_fit(data):
|
||||
features = ['age', 'height']
|
||||
|
|
@ -73,7 +68,8 @@ def test_minimizer_fit(data):
|
|||
gen.fit(X, predictions)
|
||||
transformed = gen.transform(X)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {}, 'categories': {}, 'untouched': ['age', 'height']}
|
||||
expexted_generalizations = {'ranges': {}, 'categories': {}, 'untouched': ['height', 'age']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
|
|
@ -136,7 +132,8 @@ def test_minimizer_fit_pandas(data):
|
|||
gen.fit(X, predictions)
|
||||
transformed = gen.transform(X)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {'age': []}, 'categories': {}, 'untouched': ['sex', 'height', 'ola']}
|
||||
expexted_generalizations = {'ranges': {'age': []}, 'categories': {}, 'untouched': ['ola', 'height', 'sex']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
|
|
@ -206,17 +203,113 @@ def test_minimizer_params_categorical(data):
|
|||
# Append classifier to preprocessing pipeline.
|
||||
# Now we have a full prediction pipeline.
|
||||
gen = GeneralizeToRepresentative(base_est, features=features, target_accuracy=0.5,
|
||||
categorical_features=categorical_features)
|
||||
categorical_features=categorical_features, cells=cells)
|
||||
gen.fit(X, predictions)
|
||||
transformed = gen.transform(X)
|
||||
|
||||
|
||||
def test_minimizer_fit_QI(data):
|
||||
features = ['age', 'height', 'weight']
|
||||
X = np.array([[23, 165, 70],
|
||||
[45, 158, 67],
|
||||
[56, 123, 65],
|
||||
[67, 154, 90],
|
||||
[45, 149, 67],
|
||||
[42, 166, 58],
|
||||
[73, 172, 68],
|
||||
[94, 168, 69],
|
||||
[69, 175, 80],
|
||||
[24, 181, 95],
|
||||
[18, 190, 102]])
|
||||
print(X)
|
||||
y = [1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 0]
|
||||
QI = [0, 2]
|
||||
base_est = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
base_est.fit(X, y)
|
||||
predictions = base_est.predict(X)
|
||||
|
||||
gen = GeneralizeToRepresentative(base_est, features=features, target_accuracy=0.5, features_to_minimize=QI)
|
||||
gen.fit(X, predictions)
|
||||
transformed = gen.transform(X)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {'age': []}, 'categories': {}, 'untouched': ['height', 'sex']}
|
||||
expexted_generalizations = {'ranges': {'age': [], 'weight': [67.5]}, 'categories': {}, 'untouched': ['height']}
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
assert ((np.delete(transformed, QI, axis=1) == np.delete(X, QI, axis=1)).all())
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
indexes = []
|
||||
for i in range(len(features)):
|
||||
if features[i] in modified_features:
|
||||
indexes.append(i)
|
||||
assert ((np.delete(transformed, indexes, axis=1) == np.delete(X, indexes, axis=1)).all())
|
||||
ncp = gen.ncp_
|
||||
if len(expexted_generalizations['ranges'].keys()) > 0 or len(expexted_generalizations['categories'].keys()) > 0:
|
||||
assert (ncp > 0)
|
||||
assert (((transformed[indexes]) != (X[indexes])).any())
|
||||
|
||||
|
||||
def test_minimizer_fit_pandas_QI(data):
|
||||
features = ['age', 'height', 'weight', 'sex', 'ola']
|
||||
X = [[23, 165, 65, 'f', 'aa'],
|
||||
[45, 158, 76, 'f', 'aa'],
|
||||
[56, 123, 78, 'f', 'bb'],
|
||||
[67, 154, 87, 'm', 'aa'],
|
||||
[45, 149, 45, 'f', 'bb'],
|
||||
[42, 166, 76, 'm', 'bb'],
|
||||
[73, 172, 85, 'm', 'bb'],
|
||||
[94, 168, 92, 'f', 'aa'],
|
||||
[69, 175, 95, 'm', 'aa'],
|
||||
[24, 181, 49, 'm', 'bb'],
|
||||
[18, 190, 69, 'm', 'bb']]
|
||||
|
||||
y = [1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 0]
|
||||
X = pd.DataFrame(X, columns=features)
|
||||
QI = ['age', 'weight', 'ola']
|
||||
|
||||
numeric_features = ["age", "height", "weight"]
|
||||
numeric_transformer = Pipeline(
|
||||
steps=[('imputer', SimpleImputer(strategy='constant', fill_value=0))]
|
||||
)
|
||||
|
||||
categorical_features = ["sex", "ola"]
|
||||
categorical_transformer = OneHotEncoder(handle_unknown="ignore")
|
||||
|
||||
preprocessor = ColumnTransformer(
|
||||
transformers=[
|
||||
("num", numeric_transformer, numeric_features),
|
||||
("cat", categorical_transformer, categorical_features),
|
||||
]
|
||||
)
|
||||
encoded = preprocessor.fit_transform(X)
|
||||
base_est = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
base_est.fit(encoded, y)
|
||||
predictions = base_est.predict(encoded)
|
||||
# Append classifier to preprocessing pipeline.
|
||||
# Now we have a full prediction pipeline.
|
||||
gen = GeneralizeToRepresentative(base_est, features=features, target_accuracy=0.5,
|
||||
categorical_features=categorical_features, features_to_minimize=QI)
|
||||
gen.fit(X, predictions)
|
||||
transformed = gen.transform(X)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {'age': [], 'weight': [47.0]}, 'categories': {'ola': [['bb', 'aa']]},
|
||||
'untouched': ['height', 'sex']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
assert (transformed.drop(QI, axis=1).equals(X.drop(QI, axis=1)))
|
||||
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
|
|
@ -230,24 +323,27 @@ def test_minimizer_params_categorical(data):
|
|||
def test_minimize_ndarray_iris():
|
||||
features = ['sepal length (cm)', 'sepal width (cm)', 'petal length (cm)', 'petal width (cm)']
|
||||
(x_train, y_train), _ = get_iris_dataset()
|
||||
QI = [0, 2]
|
||||
model = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
model.fit(x_train, y_train)
|
||||
pred = model.predict(x_train)
|
||||
|
||||
gen = GeneralizeToRepresentative(model, target_accuracy=0.7, features=features)
|
||||
gen = GeneralizeToRepresentative(model, target_accuracy=0.3, features=features, features_to_minimize=QI)
|
||||
gen.fit(x_train, pred)
|
||||
transformed = gen.transform(x_train)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {
|
||||
'ranges': {'sepal length (cm)': [5.0], 'sepal width (cm)': [], 'petal length (cm)': [4.950000047683716],
|
||||
'petal width (cm)': [0.800000011920929, 1.699999988079071]}, 'categories': {}, 'untouched': []}
|
||||
expexted_generalizations = {'ranges': {'sepal length (cm)': [], 'petal length (cm)': [2.449999988079071]},
|
||||
'categories': {}, 'untouched': ['petal width (cm)', 'sepal width (cm)']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
assert ((np.delete(transformed, QI, axis=1) == np.delete(x_train, QI, axis=1)).all())
|
||||
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
|
|
@ -262,70 +358,19 @@ def test_minimize_ndarray_iris():
|
|||
assert (((transformed[indexes]) != (x_train[indexes])).any())
|
||||
|
||||
|
||||
def test_minimize_pandas_nursery():
|
||||
(x_train, y_train), _ = get_nursery_dataset()
|
||||
x_train = x_train.astype(str)
|
||||
x_train.reset_index(inplace=True, drop=True)
|
||||
y_train.reset_index(inplace=True, drop=True)
|
||||
QI = ["finance", "social", "health"]
|
||||
features = ["parents", "has_nurs", "form", "children", "housing", "finance", "social", "health"]
|
||||
categorical_features = ["parents", "has_nurs", "form", "housing", "finance", "social", "health", 'children']
|
||||
numeric_features = [f for f in features if f not in categorical_features]
|
||||
numeric_transformer = Pipeline(
|
||||
steps=[('imputer', SimpleImputer(strategy='constant', fill_value=0))]
|
||||
)
|
||||
categorical_transformer = OneHotEncoder(handle_unknown="ignore", sparse=False)
|
||||
preprocessor = ColumnTransformer(
|
||||
transformers=[
|
||||
("num", numeric_transformer, numeric_features),
|
||||
("cat", categorical_transformer, categorical_features),
|
||||
]
|
||||
)
|
||||
encoded = preprocessor.fit_transform(x_train)
|
||||
base_est = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
base_est.fit(encoded, y_train)
|
||||
predictions = base_est.predict(encoded)
|
||||
|
||||
gen = GeneralizeToRepresentative(base_est, target_accuracy=0.8, features=features,
|
||||
categorical_features=categorical_features)
|
||||
gen.fit(x_train, predictions)
|
||||
transformed = gen.transform(x_train)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {}, 'categories': {'parents': [['great_pret', 'pretentious', 'usual']],
|
||||
'has_nurs': [['critical', 'less_proper', 'proper'],
|
||||
['very_crit'], ['improper']], 'form': [
|
||||
['foster', 'completed', 'complete', 'incomplete']], 'housing': [['convenient', 'less_conv', 'critical']],
|
||||
'finance': [['convenient', 'inconv']],
|
||||
'social': [['problematic', 'nonprob', 'slightly_prob']],
|
||||
'health': [['priority'], ['recommended'], ['not_recom']],
|
||||
'children': [['2', '3', '4', '1']]}, 'untouched': []}
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
assert (transformed.drop(modified_features, axis=1).equals(x_train.drop(modified_features, axis=1)))
|
||||
ncp = gen.ncp_
|
||||
if len(expexted_generalizations['ranges'].keys()) > 0 or len(expexted_generalizations['categories'].keys()) > 0:
|
||||
assert (ncp > 0)
|
||||
assert (((transformed[modified_features]).equals(x_train[modified_features])) == False)
|
||||
|
||||
|
||||
def test_minimize_pandas_adult():
|
||||
(x_train, y_train), _ = get_adult_dataset()
|
||||
x_train = x_train.head(5000)
|
||||
y_train = y_train.head(5000)
|
||||
x_train = x_train.head(1000)
|
||||
y_train = y_train.head(1000)
|
||||
|
||||
features = ['age', 'workclass', 'education-num', 'marital-status', 'occupation', 'relationship', 'race', 'sex',
|
||||
'capital-gain', 'capital-loss', 'hours-per-week', 'native-country']
|
||||
|
||||
categorical_features = ['workclass', 'marital-status', 'occupation', 'relationship', 'race', 'sex',
|
||||
'native-country']
|
||||
'hours-per-week', 'native-country']
|
||||
|
||||
QI = ['age', 'workclass', 'education-num', 'marital-status', 'occupation', 'relationship', 'race', 'sex',
|
||||
'native-country']
|
||||
|
||||
numeric_features = [f for f in features if f not in categorical_features]
|
||||
numeric_transformer = Pipeline(
|
||||
|
|
@ -344,33 +389,101 @@ def test_minimize_pandas_adult():
|
|||
base_est.fit(encoded, y_train)
|
||||
predictions = base_est.predict(encoded)
|
||||
|
||||
gen = GeneralizeToRepresentative(base_est, target_accuracy=0.8, features=features,
|
||||
categorical_features=categorical_features)
|
||||
gen = GeneralizeToRepresentative(base_est, target_accuracy=0.7, features=features,
|
||||
categorical_features=categorical_features, features_to_minimize=QI)
|
||||
gen.fit(x_train, predictions)
|
||||
transformed = gen.transform(x_train)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {
|
||||
'ranges': {'age': [20.0], 'education-num': [11.5, 12.5], 'capital-gain': [5095.5, 7139.5], 'capital-loss': [],
|
||||
'hours-per-week': []}, 'categories': {'workclass': [
|
||||
['Private', 'Without-pay', 'Self-emp-not-inc', '?', 'Federal-gov', 'Self-emp-inc', 'State-gov',
|
||||
'Local-gov']], 'marital-status': [
|
||||
['Married-civ-spouse', 'Never-married', 'Widowed', 'Married-AF-spouse', 'Separated',
|
||||
'Married-spouse-absent'], ['Divorced']], 'occupation': [
|
||||
['Transport-moving', 'Priv-house-serv', '?', 'Armed-Forces', 'Prof-specialty', 'Farming-fishing',
|
||||
'Exec-managerial', 'Machine-op-inspct', 'Other-service', 'Sales', 'Protective-serv', 'Handlers-cleaners',
|
||||
'Tech-support', 'Craft-repair', 'Adm-clerical']], 'relationship': [
|
||||
['Not-in-family', 'Own-child', 'Wife', 'Other-relative', 'Husband', 'Unmarried']], 'race': [
|
||||
['Other', 'Asian-Pac-Islander', 'Black', 'White', 'Amer-Indian-Eskimo']], 'sex': [['Male', 'Female']],
|
||||
'native-country': [
|
||||
['LatinAmerica', 'Other', 'UnitedStates', 'SouthAmerica',
|
||||
'BritishCommonwealth', 'Euro_2', 'Unknown', 'China',
|
||||
'Euro_1', 'SE_Asia']]}, 'untouched': []}
|
||||
expexted_generalizations = {'ranges': {'age': [], 'education-num': []}, 'categories': {
|
||||
'workclass': [['Self-emp-not-inc', 'Private', 'Federal-gov', 'Self-emp-inc', '?', 'Local-gov', 'State-gov']],
|
||||
'marital-status': [
|
||||
['Divorced', 'Married-AF-spouse', 'Married-spouse-absent', 'Widowed', 'Separated', 'Married-civ-spouse',
|
||||
'Never-married']], 'occupation': [
|
||||
['Tech-support', 'Priv-house-serv', 'Machine-op-inspct', 'Other-service', 'Prof-specialty', 'Adm-clerical',
|
||||
'Protective-serv', 'Handlers-cleaners', 'Transport-moving', 'Armed-Forces', '?', 'Sales',
|
||||
'Farming-fishing', 'Exec-managerial', 'Craft-repair']],
|
||||
'relationship': [['Not-in-family', 'Wife', 'Other-relative', 'Husband', 'Unmarried', 'Own-child']],
|
||||
'race': [['Asian-Pac-Islander', 'White', 'Other', 'Black', 'Amer-Indian-Eskimo']], 'sex': [['Female', 'Male']],
|
||||
'native-country': [
|
||||
['Euro_1', 'LatinAmerica', 'BritishCommonwealth', 'SouthAmerica', 'UnitedStates', 'China', 'Euro_2',
|
||||
'SE_Asia', 'Other', 'Unknown']]}, 'untouched': ['capital-loss', 'hours-per-week', 'capital-gain']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
assert (transformed.drop(QI, axis=1).equals(x_train.drop(QI, axis=1)))
|
||||
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
assert (transformed.drop(modified_features, axis=1).equals(x_train.drop(modified_features, axis=1)))
|
||||
ncp = gen.ncp_
|
||||
if len(expexted_generalizations['ranges'].keys()) > 0 or len(expexted_generalizations['categories'].keys()) > 0:
|
||||
assert (ncp > 0)
|
||||
assert (((transformed[modified_features]).equals(x_train[modified_features])) == False)
|
||||
|
||||
|
||||
def test_german_credit_pandas():
|
||||
(x_train, y_train), (x_test, y_test) = get_german_credit_dataset()
|
||||
features = ["Existing_checking_account", "Duration_in_month", "Credit_history", "Purpose", "Credit_amount",
|
||||
"Savings_account", "Present_employment_since", "Installment_rate", "Personal_status_sex", "debtors",
|
||||
"Present_residence", "Property", "Age", "Other_installment_plans", "Housing",
|
||||
"Number_of_existing_credits", "Job", "N_people_being_liable_provide_maintenance", "Telephone",
|
||||
"Foreign_worker"]
|
||||
categorical_features = ["Existing_checking_account", "Credit_history", "Purpose", "Savings_account",
|
||||
"Present_employment_since", "Personal_status_sex", "debtors", "Property",
|
||||
"Other_installment_plans", "Housing", "Job"]
|
||||
QI = ["Duration_in_month", "Credit_history", "Purpose", "debtors", "Property", "Other_installment_plans",
|
||||
"Housing", "Job"]
|
||||
|
||||
|
||||
numeric_features = [f for f in features if f not in categorical_features]
|
||||
numeric_transformer = Pipeline(
|
||||
steps=[('imputer', SimpleImputer(strategy='constant', fill_value=0))]
|
||||
)
|
||||
categorical_transformer = OneHotEncoder(handle_unknown="ignore", sparse=False)
|
||||
preprocessor = ColumnTransformer(
|
||||
transformers=[
|
||||
("num", numeric_transformer, numeric_features),
|
||||
("cat", categorical_transformer, categorical_features),
|
||||
]
|
||||
)
|
||||
encoded = preprocessor.fit_transform(x_train)
|
||||
base_est = DecisionTreeClassifier(random_state=0, min_samples_split=2,
|
||||
min_samples_leaf=1)
|
||||
base_est.fit(encoded, y_train)
|
||||
predictions = base_est.predict(encoded)
|
||||
|
||||
gen = GeneralizeToRepresentative(base_est, target_accuracy=0.7, features=features,
|
||||
categorical_features=categorical_features, features_to_minimize=QI)
|
||||
gen.fit(x_train, predictions)
|
||||
transformed = gen.transform(x_train)
|
||||
gener = gen.generalizations_
|
||||
expexted_generalizations = {'ranges': {'Duration_in_month': [31.5]},
|
||||
'categories': {'Credit_history': [['A30', 'A32', 'A31', 'A34', 'A33']], 'Purpose': [
|
||||
['A41', 'A46', 'A43', 'A40', 'A44', 'A410', 'A49', 'A45', 'A48', 'A42']],
|
||||
'debtors': [['A101', 'A102', 'A103']],
|
||||
'Property': [['A124', 'A121', 'A122', 'A123']],
|
||||
'Other_installment_plans': [['A142', 'A141', 'A143']],
|
||||
'Housing': [['A151', 'A152', 'A153']],
|
||||
'Job': [['A172', 'A171', 'A174', 'A173']]},
|
||||
'untouched': ['Installment_rate', 'Present_residence', 'Personal_status_sex',
|
||||
'Foreign_worker', 'Telephone', 'Savings_account',
|
||||
'Number_of_existing_credits', 'N_people_being_liable_provide_maintenance',
|
||||
'Age', 'Existing_checking_account', 'Credit_amount',
|
||||
'Present_employment_since']}
|
||||
|
||||
for key in expexted_generalizations['ranges']:
|
||||
assert (set(expexted_generalizations['ranges'][key]) == set(gener['ranges'][key]))
|
||||
for key in expexted_generalizations['categories']:
|
||||
assert (set([frozenset(sl) for sl in expexted_generalizations['categories'][key]]) ==
|
||||
set([frozenset(sl) for sl in gener['categories'][key]]))
|
||||
assert (set(expexted_generalizations['untouched']) == set(gener['untouched']))
|
||||
assert (transformed.drop(QI, axis=1).equals(x_train.drop(QI, axis=1)))
|
||||
|
||||
modified_features = [f for f in features if
|
||||
f in expexted_generalizations['categories'].keys() or f in expexted_generalizations[
|
||||
'ranges'].keys()]
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue