diff --git a/README.md b/README.md index 2a1818d..f66f918 100644 --- a/README.md +++ b/README.md @@ -30,7 +30,7 @@ This library is in beta, and currently not all models are supported. The library | [Ridge](sklearn_pmml_model/linear_model) | ✅2 | ✅ | ✅3 | | [Lasso](sklearn_pmml_model/linear_model) | ✅2 | ✅ | ✅3 | | [ElasticNet](sklearn_pmml_model/linear_model) | ✅2 | ✅ | ✅ | -| [Gaussian Naive Bayes](sklearn_pmml_model/naive_bayes) | ✅ | | | +| [Gaussian Naive Bayes](sklearn_pmml_model/naive_bayes) | ✅ | | ✅3 | 1 Categorical feature support using slightly modified internals, based on [scikit-learn#12866](https://github.com/scikit-learn/scikit-learn/pull/12866). diff --git a/models/tree-iris.pmml b/models/tree-iris.pmml index b8ace33..f2f2fbf 100644 --- a/models/tree-iris.pmml +++ b/models/tree-iris.pmml @@ -1,396 +1,111 @@ - +
- - 2018-06-18T14:47:30Z + + 2021-07-06T10:18:03Z
- PMMLPipeline(steps=[('classifier', DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=None, - max_features=None, max_leaf_nodes=None, - min_impurity_decrease=0.0, min_impurity_split=None, - min_samples_leaf=1, min_samples_split=2, - min_weight_fraction_leaf=0.0, presort=False, random_state=1, - splitter='best'))]) + PMMLPipeline(steps=[('classifier', DecisionTreeClassifier(random_state=1))]) - - - - - - - - - - + + + + + + + - - - - - - - - - + + + + - - - - + + + - + + + + + + + + + + + + - - - - - - - - - + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - - - - - - - - - + + + + + - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + diff --git a/sklearn_pmml_model/base.py b/sklearn_pmml_model/base.py index c50b7ac..c2dc373 100644 --- a/sklearn_pmml_model/base.py +++ b/sklearn_pmml_model/base.py @@ -1,11 +1,13 @@ from sklearn.base import BaseEstimator -from sklearn.preprocessing import LabelBinarizer +from sklearn.preprocessing import LabelBinarizer, OneHotEncoder +from sklearn.compose import ColumnTransformer from xml.etree import cElementTree as eTree from cached_property import cached_property from sklearn_pmml_model.datatypes import Category from collections import OrderedDict import datetime import numpy as np +import pandas as pd class PMMLBaseEstimator(BaseEstimator): @@ -137,13 +139,18 @@ def fit(self, x, y): raise Exception('Not supported.') def _prepare_data(self, X): - X = np.asarray(X) + pmml_features = [f for f,e in self.fields.items() if e is not self.target_field and e.tag == 'DataField'] - for column, (index, field_type) in self.field_mapping.items(): - if type(field_type) is Category and index is not None and type(X[0,index]) is str: - categories = [str(v) for v in field_type.categories] - categories += [c for c in np.unique(X[:,index]) if c not in categories] - X[:,index] = [categories.index(x) for x in X[:,index]] + if isinstance(X, pd.DataFrame): + X.columns = X.columns.map(str) + + try: + X = X[pmml_features] + except KeyError: + raise Exception('The features in the input data do not match features expected by the PMML model.') + elif X.shape[1] != len(pmml_features): + raise Exception('The number of features in provided data does not match expected number of features in the PMML. ' + 'Provide pandas.Dataframe, or provide data matching the DataFields in the PMML document.') return X @@ -258,3 +265,60 @@ def findall(element, path): if element is None: return [] return element.findall(path) + + +class OneHotEncodingMixin: + """ + Mixin class to automatically one-hot encode categorical variables. + + """ + def __init__(self): + # Setup a column transformer to encode categorical variables + target = self.target_field.get('name') + fields = [field for name, field in self.fields.items() if name != target] + + def encoder_for(field): + if field.get('optype') != 'categorical': + return 'passthrough' + + encoder = OneHotEncoder() + encoder.categories_ = np.array([self.field_mapping[field.get('name')][1].categories]) + encoder.drop_idx_ = np.array([None for x in encoder.categories_]) + encoder._legacy_mode = False + return encoder + + transformer = ColumnTransformer( + transformers=[ + (field.get('name'), encoder_for(field), [self.field_mapping[field.get('name')][0]]) + for field in fields + if field.tag == 'DataField' + ] + ) + + X = np.array([[0 for field in fields if field.tag == "DataField"]]) + transformer._validate_transformers() + transformer._validate_column_callables(X) + transformer._validate_remainder(X) + transformer.transformers_ = transformer.transformers + transformer.sparse_output_ = False + transformer._feature_names_in = None + + self.transformer = transformer + + def _prepare_data(self, X): + X = super()._prepare_data(X) + return self.transformer.transform(X) + + +class IntegerEncodingMixin: + def _prepare_data(self, X): + X = super()._prepare_data(X) + X = np.asarray(X) + + for column, (index, field_type) in self.field_mapping.items(): + if type(field_type) is Category and index is not None and type(X[0, index]) is str: + categories = [str(v) for v in field_type.categories] + categories += [c for c in np.unique(X[:, index]) if c not in categories] + X[:, index] = [categories.index(x) for x in X[:, index]] + + return X \ No newline at end of file diff --git a/sklearn_pmml_model/ensemble/forest.py b/sklearn_pmml_model/ensemble/forest.py index 771e1fe..e29f37b 100644 --- a/sklearn_pmml_model/ensemble/forest.py +++ b/sklearn_pmml_model/ensemble/forest.py @@ -1,11 +1,11 @@ import numpy as np import warnings from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor -from sklearn_pmml_model.base import PMMLBaseClassifier, PMMLBaseRegressor +from sklearn_pmml_model.base import PMMLBaseClassifier, PMMLBaseRegressor, IntegerEncodingMixin from sklearn_pmml_model.tree import get_tree -class PMMLForestClassifier(PMMLBaseClassifier, RandomForestClassifier): +class PMMLForestClassifier(IntegerEncodingMixin, PMMLBaseClassifier, RandomForestClassifier): """ A random forest classifier. @@ -92,7 +92,7 @@ def _more_tags(self): return RandomForestClassifier._more_tags(self) -class PMMLForestRegressor(PMMLBaseRegressor, RandomForestRegressor): +class PMMLForestRegressor(IntegerEncodingMixin, PMMLBaseRegressor, RandomForestRegressor): """ A random forest regressor. diff --git a/sklearn_pmml_model/ensemble/gb.py b/sklearn_pmml_model/ensemble/gb.py index a1ab5ee..8cdd4cd 100644 --- a/sklearn_pmml_model/ensemble/gb.py +++ b/sklearn_pmml_model/ensemble/gb.py @@ -3,13 +3,13 @@ import numpy as np from sklearn.tree import DecisionTreeRegressor from sklearn.ensemble import GradientBoostingClassifier, GradientBoostingRegressor, _gb_losses -from sklearn_pmml_model.base import PMMLBaseClassifier, PMMLBaseRegressor +from sklearn_pmml_model.base import PMMLBaseClassifier, PMMLBaseRegressor, IntegerEncodingMixin from sklearn_pmml_model.tree import get_tree from scipy.special import expit from ._gradient_boosting import predict_stages -class PMMLGradientBoostingClassifier(PMMLBaseClassifier, GradientBoostingClassifier, ABC): +class PMMLGradientBoostingClassifier(IntegerEncodingMixin, PMMLBaseClassifier, GradientBoostingClassifier, ABC): """ Gradient Boosting for classification. @@ -135,7 +135,7 @@ def _more_tags(self): return GradientBoostingClassifier._more_tags(self) -class PMMLGradientBoostingRegressor(PMMLBaseRegressor, GradientBoostingRegressor, ABC): +class PMMLGradientBoostingRegressor(IntegerEncodingMixin, PMMLBaseRegressor, GradientBoostingRegressor, ABC): """ Gradient Boosting for regression. diff --git a/sklearn_pmml_model/linear_model/base.py b/sklearn_pmml_model/linear_model/base.py index 75870fd..6211a68 100644 --- a/sklearn_pmml_model/linear_model/base.py +++ b/sklearn_pmml_model/linear_model/base.py @@ -1,107 +1,9 @@ -from sklearn_pmml_model.base import PMMLBaseRegressor, PMMLBaseClassifier -from sklearn.preprocessing import OneHotEncoder -from sklearn.compose import ColumnTransformer +from sklearn_pmml_model.base import PMMLBaseRegressor, PMMLBaseClassifier, OneHotEncodingMixin import numpy as np from itertools import chain -class PMMLLinearModel(PMMLBaseRegressor): - """ - Abstract class for linear models. - - """ - def __init__(self, pmml): - PMMLBaseRegressor.__init__(self, pmml) - - # Setup a column transformer to deal with categorical variables - target = self.target_field.get('name') - fields = [field for name, field in self.fields.items() if name != target] - - def encoder_for(field): - if field.get('optype') != 'categorical': - return 'passthrough' - - encoder = OneHotEncoder() - encoder.categories_ = np.array([self.field_mapping[field.get('name')][1].categories]) - encoder.drop_idx_ = np.array([None for x in encoder.categories_]) - encoder._legacy_mode = False - return encoder - - transformer = ColumnTransformer( - transformers=[ - (field.get('name'), encoder_for(field), [self.field_mapping[field.get('name')][0]]) - for field in fields - if field.tag == 'DataField' - ] - ) - - X = np.array([[0 for field in fields if field.tag == "DataField"]]) - transformer._validate_transformers() - transformer._validate_column_callables(X) - transformer._validate_remainder(X) - transformer.transformers_ = transformer.transformers - transformer.sparse_output_ = False - transformer._feature_names_in = None - - self.transformer = transformer - - def _prepare_data(self, X): - """ - Overrides the default data preparation operation by one-hot encoding - categorical variables. - """ - return self.transformer.transform(X) - - -class PMMLLinearClassifier(PMMLBaseClassifier): - """ - Abstract class for linear models. - - """ - def __init__(self, pmml): - PMMLBaseClassifier.__init__(self, pmml) - - # Setup a column transformer to deal with categorical variables - target = self.target_field.get('name') - fields = [field for name, field in self.fields.items() if name != target] - - def encoder_for(field): - if field.get('optype') != 'categorical': - return 'passthrough' - - encoder = OneHotEncoder() - encoder.categories_ = np.array([self.field_mapping[field.get('name')][1].categories]) - encoder.drop_idx_ = np.array([None for x in encoder.categories_]) - encoder._legacy_mode = False - return encoder - - transformer = ColumnTransformer( - transformers=[ - (field.get('name'), encoder_for(field), [self.field_mapping[field.get('name')][0]]) - for field in fields - if field.tag == 'DataField' - ] - ) - - X = np.array([[0 for field in fields if field.tag == "DataField"]]) - transformer._validate_transformers() - transformer._validate_column_callables(X) - transformer._validate_remainder(X) - transformer.transformers_ = transformer.transformers - transformer.sparse_output_ = False - transformer._feature_names_in = None - - self.transformer = transformer - - def _prepare_data(self, X): - """ - Overrides the default data preparation operation by one-hot encoding - categorical variables. - """ - return self.transformer.transform(X) - - -class PMMLGeneralizedLinearRegressor(PMMLLinearModel): +class PMMLGeneralizedLinearRegressor(OneHotEncodingMixin, PMMLBaseRegressor): """ Abstract class for Generalized Linear Models (GLMs). @@ -122,7 +24,8 @@ class PMMLGeneralizedLinearRegressor(PMMLLinearModel): """ def __init__(self, pmml): - PMMLLinearModel.__init__(self, pmml) + PMMLBaseRegressor.__init__(self, pmml) + OneHotEncodingMixin.__init__(self) # Import coefficients and intercepts model = self.root.find('GeneralRegressionModel') @@ -134,7 +37,7 @@ def __init__(self, pmml): self.intercept_ = _get_intercept(model) -class PMMLGeneralizedLinearClassifier(PMMLLinearClassifier): +class PMMLGeneralizedLinearClassifier(OneHotEncodingMixin, PMMLBaseClassifier): """ Abstract class for Generalized Linear Models (GLMs). @@ -155,7 +58,8 @@ class PMMLGeneralizedLinearClassifier(PMMLLinearClassifier): """ def __init__(self, pmml): - PMMLLinearClassifier.__init__(self, pmml) + PMMLBaseClassifier.__init__(self, pmml) + OneHotEncodingMixin.__init__(self) # Import coefficients and intercepts model = self.root.find('GeneralRegressionModel') diff --git a/sklearn_pmml_model/linear_model/implementations.py b/sklearn_pmml_model/linear_model/implementations.py index 9109899..457ee67 100644 --- a/sklearn_pmml_model/linear_model/implementations.py +++ b/sklearn_pmml_model/linear_model/implementations.py @@ -1,11 +1,11 @@ from sklearn.linear_model import LinearRegression, Ridge, RidgeClassifier, Lasso, ElasticNet, LogisticRegression -from sklearn_pmml_model.linear_model.base import PMMLLinearModel, PMMLLinearClassifier, PMMLGeneralizedLinearRegressor,\ - PMMLGeneralizedLinearClassifier +from sklearn_pmml_model.base import PMMLBaseRegressor, PMMLBaseClassifier, OneHotEncodingMixin +from sklearn_pmml_model.linear_model.base import PMMLGeneralizedLinearRegressor, PMMLGeneralizedLinearClassifier from itertools import chain import numpy as np -class PMMLLinearRegression(PMMLLinearModel, LinearRegression): +class PMMLLinearRegression(OneHotEncodingMixin, PMMLBaseRegressor, LinearRegression): """ Ordinary least squares Linear Regression. @@ -25,7 +25,8 @@ class PMMLLinearRegression(PMMLLinearModel, LinearRegression): """ def __init__(self, pmml): - PMMLLinearModel.__init__(self, pmml) + PMMLBaseRegressor.__init__(self, pmml) + OneHotEncodingMixin.__init__(self) # Import coefficients and intercepts model = self.root.find('RegressionModel') @@ -51,13 +52,13 @@ def __init__(self, pmml): self.intercept_ = self.intercept_[0] def fit(self, x, y): - return PMMLLinearModel.fit(self, x, y) + return PMMLBaseRegressor.fit(self, x, y) def _more_tags(self): return LinearRegression._more_tags(self) -class PMMLLogisticRegression(PMMLLinearClassifier, LogisticRegression): +class PMMLLogisticRegression(OneHotEncodingMixin, PMMLBaseClassifier, LogisticRegression): """ Logistic Regression (aka logit, MaxEnt) classifier. @@ -77,7 +78,8 @@ class PMMLLogisticRegression(PMMLLinearClassifier, LogisticRegression): """ def __init__(self, pmml): - PMMLLinearClassifier.__init__(self, pmml) + PMMLBaseClassifier.__init__(self, pmml) + OneHotEncodingMixin.__init__(self) # Import coefficients and intercepts model = self.root.find('RegressionModel') @@ -111,7 +113,7 @@ def __init__(self, pmml): self.solver = 'lbfgs' def fit(self, x, y): - return PMMLLinearClassifier.fit(self, x, y) + return PMMLBaseClassifier.fit(self, x, y) def _more_tags(self): return LogisticRegression._more_tags(self) diff --git a/sklearn_pmml_model/naive_bayes/implementations.py b/sklearn_pmml_model/naive_bayes/implementations.py index a94d7c5..5cc91d0 100644 --- a/sklearn_pmml_model/naive_bayes/implementations.py +++ b/sklearn_pmml_model/naive_bayes/implementations.py @@ -1,10 +1,10 @@ -from sklearn_pmml_model.base import PMMLBaseClassifier +from sklearn_pmml_model.base import PMMLBaseClassifier, OneHotEncodingMixin from sklearn.naive_bayes import GaussianNB import numpy as np from itertools import chain -class PMMLGaussianNB(PMMLBaseClassifier, GaussianNB): +class PMMLGaussianNB(OneHotEncodingMixin, PMMLBaseClassifier, GaussianNB): """ Gaussian Naive Bayes (GaussianNB) @@ -26,6 +26,7 @@ class PMMLGaussianNB(PMMLBaseClassifier, GaussianNB): """ def __init__(self, pmml): PMMLBaseClassifier.__init__(self, pmml) + OneHotEncodingMixin.__init__(self) model = self.root.find('NaiveBayesModel') diff --git a/tests/naive_bayes/test_naive_bayes.py b/tests/naive_bayes/test_naive_bayes.py index 7376ea4..e31e513 100644 --- a/tests/naive_bayes/test_naive_bayes.py +++ b/tests/naive_bayes/test_naive_bayes.py @@ -74,15 +74,16 @@ class TestGaussianNBIntegration(TestCase): def setUp(self): df = pd.read_csv(path.join(BASE_DIR, '../models/categorical-test.csv')) Xte = df.iloc[:, 1:] - Xte = pd.get_dummies(Xte, prefix_sep='') + Xenc = pd.get_dummies(Xte, prefix_sep='') yte = df.iloc[:, 0] self.test = (Xte, yte) + self.enc = (Xenc, yte) pmml = path.join(BASE_DIR, '../models/nb-cat-pima.pmml') self.clf = PMMLGaussianNB(pmml) self.ref = GaussianNB() - self.ref.fit(Xte, yte) + self.ref.fit(Xenc, yte) def test_predict_proba(self): Xte, _ = self.test @@ -110,7 +111,7 @@ def test_sklearn2pmml(self): pipeline = PMMLPipeline([ ("classifier", self.ref) ]) - pipeline.fit(self.test[0], self.test[1]) + pipeline.fit(self.enc[0], self.enc[1]) sklearn2pmml(pipeline, "gnb-sklearn2pmml.pmml", with_repr = True) try: @@ -118,10 +119,10 @@ def test_sklearn2pmml(self): model = PMMLGaussianNB(pmml='gnb-sklearn2pmml.pmml') # Verify classification - Xte, _ = self.test - assert np.array_equal( - self.ref.predict_proba(Xte), - model.predict_proba(Xte) + Xenc, _ = self.enc + assert np.allclose( + self.ref.predict_proba(Xenc), + model.predict_proba(Xenc) ) finally: @@ -152,7 +153,7 @@ def test_sklearn2pmml(self): # Verify classification Xte, _ = self.test - assert np.array_equal( + assert np.allclose( self.ref.predict_proba(Xte), model.predict_proba(Xte) ) diff --git a/tests/test_base.py b/tests/test_base.py index 22f3436..92b044b 100644 --- a/tests/test_base.py +++ b/tests/test_base.py @@ -221,3 +221,102 @@ def test_fit_exception(self): clf.fit(X, y) assert str(cm.exception) == "Not supported." + + def test_prepare_data_removes_unused_columns(self): + clf = PMMLBaseEstimator(pmml=StringIO(""" + + + + + + + + + + + + + + """)) + + X = pd.DataFrame(data=[[1, 2], [3, 4], [5, 6]], columns=["test1", "test2"]) + result = clf._prepare_data(X) + + assert list(X.columns) == ["test1", "test2"] + assert list(result.columns) == ["test1"] + + def test_prepare_data_reorders_columns(self): + clf = PMMLBaseEstimator(pmml=StringIO(""" + + + + + + + + + + + + + + + """)) + + X = pd.DataFrame(data=[[1, 2], [3, 4], [5, 6]], columns=["test1", "test2"]) + result = clf._prepare_data(X) + + assert list(X.columns) == ["test1", "test2"] + assert list(result.columns) == ["test2", "test1"] + + def test_prepare_data_exception_mismatch_columns_numpy(self): + clf = PMMLBaseEstimator(pmml=StringIO(""" + + + + + + + + + + + + + + """)) + + X = pd.DataFrame(data=[[1, 2], [3, 4], [5, 6]], columns=["test1", "test2"]) + + with self.assertRaises(Exception) as cm: + clf._prepare_data(np.asanyarray(X)) + + assert str(cm.exception) == "The number of features in provided data does not match expected number of features " \ + "in the PMML. Provide pandas.Dataframe, or provide data matching the DataFields in " \ + "the PMML document." + + def test_prepare_data_exception_mismatch_columns_pandas(self): + clf = PMMLBaseEstimator(pmml=StringIO(""" + + + + + + + + + + + + + + + """)) + + X = pd.DataFrame(data=[[1, 2], [3, 4], [5, 6]], columns=["Test_1", "Test_2"]) + + with self.assertRaises(Exception) as cm: + clf._prepare_data(X) + + assert str(cm.exception) == "The features in the input data do not match features expected by the PMML model." + diff --git a/tests/tree/test_tree.py b/tests/tree/test_tree.py index dd744dd..d2ffde8 100644 --- a/tests/tree/test_tree.py +++ b/tests/tree/test_tree.py @@ -88,15 +88,12 @@ def test_more_tags(self): class TestIrisTreeIntegration(TestCase): def setUp(self): pair = [0, 1] - data = load_iris() + data = load_iris(as_frame=True) - X = pd.DataFrame(data.data[:, pair]) - X.columns = np.array(data.feature_names)[pair] - y = pd.Series(np.array(data.target_names)[data.target]) + X = data.data + y = data.target y.name = "Class" - X, Xte, y, yte = train_test_split(X, y, test_size=0.33, random_state=123) - self.test = (Xte, yte) - self.train = (X, y) + self.test = (X, y) pmml = path.join(BASE_DIR, '../models/tree-iris.pmml') self.clf = PMMLTreeClassifier(pmml=pmml) @@ -122,7 +119,7 @@ def test_sklearn2pmml(self): pipeline = PMMLPipeline([ ("classifier", self.ref) ]) - pipeline.fit(self.train[0], self.train[1]) + pipeline.fit(self.test[0], self.test[1]) sklearn2pmml(pipeline, "tree-sklearn2pmml.pmml", with_repr = True) try: @@ -145,9 +142,6 @@ class TestDigitsTreeIntegration(TestCase): def setUp(self): data = load_digits() - self.columns = [2, 3, 4, 5, 6, 7, 9, 10, 13, 14, 17, 18, 19, 20, 21, 25, 26, - 27, 28, 29, 30, 33, 34, 35, 36, 37, 38, 41, 42, 43, 45, 46, - 50, 51, 52, 53, 54, 55, 57, 58, 59, 60, 61, 62, 63] X = pd.DataFrame(data.data) y = pd.Series(np.array(data.target_names)[data.target]) y.name = "Class" @@ -161,19 +155,19 @@ def test_predict(self): Xte, _ = self.test assert np.array_equal( self.ref.predict(Xte), - self.clf.predict(Xte[self.columns]).astype(np.int64) + self.clf.predict(Xte) ) def test_predict_proba(self): Xte, _ = self.test assert np.array_equal( self.ref.predict_proba(Xte), - self.clf.predict_proba(Xte[self.columns]) + self.clf.predict_proba(Xte) ) def test_score(self): Xte, yte = self.test - assert self.ref.score(Xte, yte) == self.clf.score(Xte[self.columns], yte) + assert self.ref.score(Xte, yte) == self.clf.score(Xte, yte) class TestCategoricalTreeIntegration(TestCase):