diff --git a/backport-CVE-2024-5206.patch b/backport-CVE-2024-5206.patch new file mode 100644 index 0000000000000000000000000000000000000000..fb0a8e82027dcdcca6f54652ef17bfbd7f6e35ba --- /dev/null +++ b/backport-CVE-2024-5206.patch @@ -0,0 +1,235 @@ +From 70ca21f106b603b611da73012c9ade7cd8e438b8 Mon Sep 17 00:00:00 2001 +From: Olivier Grisel +Date: Mon, 22 Apr 2024 15:10:46 +0200 +Subject: [PATCH] FIX remove the computed stop_words_ attribute of text + vectorizer (#28823) + +--- + doc/whats_new/v1.4.rst | 18 ++++++++ + sklearn/feature_extraction/tests/test_text.py | 42 ------------------- + sklearn/feature_extraction/text.py | 36 +--------------- + 3 files changed, 20 insertions(+), 76 deletions(-) + +diff --git a/doc/whats_new/v1.4.rst b/doc/whats_new/v1.4.rst +index ad3cc40..321db3b 100644 +--- a/doc/whats_new/v1.4.rst ++++ b/doc/whats_new/v1.4.rst +@@ -14,6 +14,24 @@ For a short description of the main highlights of the release, please refer to + + .. include:: changelog_legend.inc + ++Security ++-------- ++ ++- |Fix| :class:`feature_extraction.text.CountVectorizer` and ++ :class:`feature_extraction.text.TfidfVectorizer` no longer store discarded ++ tokens from the training set in their `stop_words_` attribute. This attribute ++ would hold too frequent (above `max_df`) but also too rare tokens (below ++ `min_df`). This fixes a potential security issue (data leak) if the discarded ++ rare tokens hold sensitive information from the training set without the ++ model developer's knowledge. ++ ++ Note: users of those classes are encouraged to either retrain their pipelines ++ with the new scikit-learn version or to manually clear the `stop_words_` ++ attribute from previously trained instances of those transformers. This ++ attribute was designed only for model inspection purposes and has no impact ++ on the behavior of the transformers. ++ :pr:`28823` by :user:`Olivier Grisel `. ++ + Changed models + -------------- + +diff --git a/sklearn/feature_extraction/tests/test_text.py b/sklearn/feature_extraction/tests/test_text.py +index 7c7cac8..b784716 100644 +--- a/sklearn/feature_extraction/tests/test_text.py ++++ b/sklearn/feature_extraction/tests/test_text.py +@@ -757,21 +757,11 @@ def test_feature_names(): + @pytest.mark.parametrize("Vectorizer", (CountVectorizer, TfidfVectorizer)) + def test_vectorizer_max_features(Vectorizer): + expected_vocabulary = {"burger", "beer", "salad", "pizza"} +- expected_stop_words = { +- "celeri", +- "tomato", +- "copyright", +- "coke", +- "sparkling", +- "water", +- "the", +- } + + # test bounded number of extracted features + vectorizer = Vectorizer(max_df=0.6, max_features=4) + vectorizer.fit(ALL_FOOD_DOCS) + assert set(vectorizer.vocabulary_) == expected_vocabulary +- assert vectorizer.stop_words_ == expected_stop_words + + + def test_count_vectorizer_max_features(): +@@ -806,21 +796,16 @@ def test_vectorizer_max_df(): + vect.fit(test_data) + assert "a" in vect.vocabulary_.keys() + assert len(vect.vocabulary_.keys()) == 6 +- assert len(vect.stop_words_) == 0 + + vect.max_df = 0.5 # 0.5 * 3 documents -> max_doc_count == 1.5 + vect.fit(test_data) + assert "a" not in vect.vocabulary_.keys() # {ae} ignored + assert len(vect.vocabulary_.keys()) == 4 # {bcdt} remain +- assert "a" in vect.stop_words_ +- assert len(vect.stop_words_) == 2 + + vect.max_df = 1 + vect.fit(test_data) + assert "a" not in vect.vocabulary_.keys() # {ae} ignored + assert len(vect.vocabulary_.keys()) == 4 # {bcdt} remain +- assert "a" in vect.stop_words_ +- assert len(vect.stop_words_) == 2 + + + def test_vectorizer_min_df(): +@@ -829,21 +814,16 @@ def test_vectorizer_min_df(): + vect.fit(test_data) + assert "a" in vect.vocabulary_.keys() + assert len(vect.vocabulary_.keys()) == 6 +- assert len(vect.stop_words_) == 0 + + vect.min_df = 2 + vect.fit(test_data) + assert "c" not in vect.vocabulary_.keys() # {bcdt} ignored + assert len(vect.vocabulary_.keys()) == 2 # {ae} remain +- assert "c" in vect.stop_words_ +- assert len(vect.stop_words_) == 4 + + vect.min_df = 0.8 # 0.8 * 3 documents -> min_doc_count == 2.4 + vect.fit(test_data) + assert "c" not in vect.vocabulary_.keys() # {bcdet} ignored + assert len(vect.vocabulary_.keys()) == 1 # {a} remains +- assert "c" in vect.stop_words_ +- assert len(vect.stop_words_) == 5 + + + def test_count_binary_occurrences(): +@@ -1156,28 +1136,6 @@ def test_countvectorizer_vocab_dicts_when_pickling(): + ) + + +-def test_stop_words_removal(): +- # Ensure that deleting the stop_words_ attribute doesn't affect transform +- +- fitted_vectorizers = ( +- TfidfVectorizer().fit(JUNK_FOOD_DOCS), +- CountVectorizer(preprocessor=strip_tags).fit(JUNK_FOOD_DOCS), +- CountVectorizer(strip_accents=strip_eacute).fit(JUNK_FOOD_DOCS), +- ) +- +- for vect in fitted_vectorizers: +- vect_transform = vect.transform(JUNK_FOOD_DOCS).toarray() +- +- vect.stop_words_ = None +- stop_None_transform = vect.transform(JUNK_FOOD_DOCS).toarray() +- +- delattr(vect, "stop_words_") +- stop_del_transform = vect.transform(JUNK_FOOD_DOCS).toarray() +- +- assert_array_equal(stop_None_transform, vect_transform) +- assert_array_equal(stop_del_transform, vect_transform) +- +- + def test_pickling_transformer(): + X = CountVectorizer().fit_transform(JUNK_FOOD_DOCS) + orig = TfidfTransformer().fit(X) +diff --git a/sklearn/feature_extraction/text.py b/sklearn/feature_extraction/text.py +index 29104c2..e9727ae 100644 +--- a/sklearn/feature_extraction/text.py ++++ b/sklearn/feature_extraction/text.py +@@ -1081,15 +1081,6 @@ class CountVectorizer(_VectorizerMixin, BaseEstimator): + True if a fixed vocabulary of term to indices mapping + is provided by the user. + +- stop_words_ : set +- Terms that were ignored because they either: +- +- - occurred in too many documents (`max_df`) +- - occurred in too few documents (`min_df`) +- - were cut off by feature selection (`max_features`). +- +- This is only available if no vocabulary was given. +- + See Also + -------- + HashingVectorizer : Convert a collection of text documents to a +@@ -1098,12 +1089,6 @@ class CountVectorizer(_VectorizerMixin, BaseEstimator): + TfidfVectorizer : Convert a collection of raw documents to a matrix + of TF-IDF features. + +- Notes +- ----- +- The ``stop_words_`` attribute can get large and increase the model size +- when pickling. This attribute is provided only for introspection and can +- be safely removed using delattr or set to None before pickling. +- + Examples + -------- + >>> from sklearn.feature_extraction.text import CountVectorizer +@@ -1242,19 +1227,17 @@ class CountVectorizer(_VectorizerMixin, BaseEstimator): + mask = new_mask + + new_indices = np.cumsum(mask) - 1 # maps old indices to new +- removed_terms = set() + for term, old_index in list(vocabulary.items()): + if mask[old_index]: + vocabulary[term] = new_indices[old_index] + else: + del vocabulary[term] +- removed_terms.add(term) + kept_indices = np.where(mask)[0] + if len(kept_indices) == 0: + raise ValueError( + "After pruning, no terms remain. Try a lower min_df or a higher max_df." + ) +- return X[:, kept_indices], removed_terms ++ return X[:, kept_indices] + + def _count_vocab(self, raw_documents, fixed_vocab): + """Create sparse feature matrix, and vocabulary where fixed_vocab=False""" +@@ -1399,7 +1382,7 @@ class CountVectorizer(_VectorizerMixin, BaseEstimator): + raise ValueError("max_df corresponds to < documents than min_df") + if max_features is not None: + X = self._sort_features(X, vocabulary) +- X, self.stop_words_ = self._limit_features( ++ X = self._limit_features( + X, vocabulary, max_doc_count, min_doc_count, max_features + ) + if max_features is None: +@@ -1932,15 +1915,6 @@ class TfidfVectorizer(CountVectorizer): + The inverse document frequency (IDF) vector; only defined + if ``use_idf`` is True. + +- stop_words_ : set +- Terms that were ignored because they either: +- +- - occurred in too many documents (`max_df`) +- - occurred in too few documents (`min_df`) +- - were cut off by feature selection (`max_features`). +- +- This is only available if no vocabulary was given. +- + See Also + -------- + CountVectorizer : Transforms text into a sparse matrix of n-gram counts. +@@ -1948,12 +1922,6 @@ class TfidfVectorizer(CountVectorizer): + TfidfTransformer : Performs the TF-IDF transformation from a provided + matrix of counts. + +- Notes +- ----- +- The ``stop_words_`` attribute can get large and increase the model size +- when pickling. This attribute is provided only for introspection and can +- be safely removed using delattr or set to None before pickling. +- + Examples + -------- + >>> from sklearn.feature_extraction.text import TfidfVectorizer +-- +2.27.0 + diff --git a/python-scikit-learn.spec b/python-scikit-learn.spec index f24f42cdb3d31cd45f66e5e7b1bbe3250b53a26d..36cd388d5fc19e542ddd7cff40f3c335fa150671 100644 --- a/python-scikit-learn.spec +++ b/python-scikit-learn.spec @@ -3,10 +3,11 @@ Name: python-scikit-learn Summary: A Python module for machine learning built on top of SciPy Version: 1.4.0 -Release: 1 +Release: 2 License: BSD URL: https://scikit-learn.org/stable/ Source0: https://files.pythonhosted.org/packages/source/s/scikit-learn/scikit-learn-%{version}.tar.gz +Patch3000: backport-CVE-2024-5206.patch %global _description\ scikit-learn is a Python module for machine learning built on top of SciPy\ @@ -43,6 +44,12 @@ CFLAGS="$RPM_OPT_FLAGS -s" %{python3_sitearch}/scikit_learn-%{version}.dist-info/ %changelog +* Fri Jun 07 2024 xuchenchen - 1.4.0-2 +- Type:CVES +- ID:CVE-2024-5206 +- SUG:NA +- DESC:fix CVE-2024-5206 + * Fri Mar 08 2024 jiangxinyu - 1.4.0-1 - Update package to version 1.4.0