# -*- coding: utf-8 -*- # Authors: Olivier Grisel # Mathieu Blondel # Lars Buitinck # Robert Layton # Jochen Wersdörfer # Roman Sinayev # # License: BSD 3 clause """ The :mod:`sklearn.feature_extraction.text` submodule gathers utilities to build feature vectors from text documents. """ from __future__ import unicode_literals import array from collections import Mapping, defaultdict import numbers from operator import itemgetter import re import unicodedata import numpy as np import scipy.sparse as sp from ..base import BaseEstimator, TransformerMixin from ..externals import six from ..externals.six.moves import xrange from ..preprocessing import normalize from .hashing import FeatureHasher from .stop_words import ENGLISH_STOP_WORDS from ..utils.validation import check_is_fitted __all__ = ['CountVectorizer', 'ENGLISH_STOP_WORDS', 'TfidfTransformer', 'TfidfVectorizer', 'strip_accents_ascii', 'strip_accents_unicode', 'strip_tags'] def strip_accents_unicode(s): """Transform accentuated unicode symbols into their simple counterpart Warning: the python-level loop and join operations make this implementation 20 times slower than the strip_accents_ascii basic normalization. See also -------- strip_accents_ascii Remove accentuated char for any unicode symbol that has a direct ASCII equivalent. """ normalized = unicodedata.normalize('NFKD', s) if normalized == s: return s else: return ''.join([c for c in normalized if not unicodedata.combining(c)]) def strip_accents_ascii(s): """Transform accentuated unicode symbols into ascii or nothing Warning: this solution is only suited for languages that have a direct transliteration to ASCII symbols. See also -------- strip_accents_unicode Remove accentuated char for any unicode symbol. """ nkfd_form = unicodedata.normalize('NFKD', s) return nkfd_form.encode('ASCII', 'ignore').decode('ASCII') def strip_tags(s): """Basic regexp based HTML / XML tag stripper function For serious HTML/XML preprocessing you should rather use an external library such as lxml or BeautifulSoup. """ return re.compile(r"<([^>]+)>", flags=re.UNICODE).sub(" ", s) def _check_stop_list(stop): if stop == "english": return ENGLISH_STOP_WORDS elif isinstance(stop, six.string_types): raise ValueError("not a built-in stop list: %s" % stop) elif stop is None: return None else: # assume it's a collection return frozenset(stop) class VectorizerMixin(object): """Provides common code for text vectorizers (tokenization logic).""" _white_spaces = re.compile(r"\s\s+") def decode(self, doc): """Decode the input into a string of unicode symbols The decoding strategy depends on the vectorizer parameters. """ if self.input == 'filename': with open(doc, 'rb') as fh: doc = fh.read() elif self.input == 'file': doc = doc.read() if isinstance(doc, bytes): doc = doc.decode(self.encoding, self.decode_error) if doc is np.nan: raise ValueError("np.nan is an invalid document, expected byte or " "unicode string.") return doc def _word_ngrams(self, tokens, stop_words=None): """Turn tokens into a sequence of n-grams after stop words filtering""" # handle stop words if stop_words is not None: tokens = [w for w in tokens if w not in stop_words] # handle token n-grams min_n, max_n = self.ngram_range if max_n != 1: original_tokens = tokens if min_n == 1: # no need to do any slicing for unigrams # just iterate through the original tokens tokens = list(original_tokens) min_n += 1 else: tokens = [] n_original_tokens = len(original_tokens) # bind method outside of loop to reduce overhead tokens_append = tokens.append space_join = " ".join for n in xrange(min_n, min(max_n + 1, n_original_tokens + 1)): for i in xrange(n_original_tokens - n + 1): tokens_append(space_join(original_tokens[i: i + n])) return tokens def _char_ngrams(self, text_document): """Tokenize text_document into a sequence of character n-grams""" # normalize white spaces text_document = self._white_spaces.sub(" ", text_document) text_len = len(text_document) min_n, max_n = self.ngram_range if min_n == 1: # no need to do any slicing for unigrams # iterate through the string ngrams = list(text_document) min_n += 1 else: ngrams = [] # bind method outside of loop to reduce overhead ngrams_append = ngrams.append for n in xrange(min_n, min(max_n + 1, text_len + 1)): for i in xrange(text_len - n + 1): ngrams_append(text_document[i: i + n]) return ngrams def _char_wb_ngrams(self, text_document): """Whitespace sensitive char-n-gram tokenization. Tokenize text_document into a sequence of character n-grams operating only inside word boundaries. n-grams at the edges of words are padded with space.""" # normalize white spaces text_document = self._white_spaces.sub(" ", text_document) min_n, max_n = self.ngram_range ngrams = [] # bind method outside of loop to reduce overhead ngrams_append = ngrams.append for w in text_document.split(): w = ' ' + w + ' ' w_len = len(w) for n in xrange(min_n, max_n + 1): offset = 0 ngrams_append(w[offset:offset + n]) while offset + n < w_len: offset += 1 ngrams_append(w[offset:offset + n]) if offset == 0: # count a short word (w_len < n) only once break return ngrams def build_preprocessor(self): """Return a function to preprocess the text before tokenization""" if self.preprocessor is not None: return self.preprocessor # unfortunately python functools package does not have an efficient # `compose` function that would have allowed us to chain a dynamic # number of functions. However the cost of a lambda call is a few # hundreds of nanoseconds which is negligible when compared to the # cost of tokenizing a string of 1000 chars for instance. noop = lambda x: x # accent stripping if not self.strip_accents: strip_accents = noop elif callable(self.strip_accents): strip_accents = self.strip_accents elif self.strip_accents == 'ascii': strip_accents = strip_accents_ascii elif self.strip_accents == 'unicode': strip_accents = strip_accents_unicode else: raise ValueError('Invalid value for "strip_accents": %s' % self.strip_accents) if self.lowercase: return lambda x: strip_accents(x.lower()) else: return strip_accents def build_tokenizer(self): """Return a function that splits a string into a sequence of tokens""" if self.tokenizer is not None: return self.tokenizer token_pattern = re.compile(self.token_pattern) return lambda doc: token_pattern.findall(doc) def get_stop_words(self): """Build or fetch the effective stop words list""" return _check_stop_list(self.stop_words) def build_analyzer(self): """Return a callable that handles preprocessing and tokenization""" if callable(self.analyzer): return self.analyzer preprocess = self.build_preprocessor() if self.analyzer == 'char': return lambda doc: self._char_ngrams(preprocess(self.decode(doc))) elif self.analyzer == 'char_wb': return lambda doc: self._char_wb_ngrams( preprocess(self.decode(doc))) elif self.analyzer == 'word': stop_words = self.get_stop_words() tokenize = self.build_tokenizer() return lambda doc: self._word_ngrams( tokenize(preprocess(self.decode(doc))), stop_words) else: raise ValueError('%s is not a valid tokenization scheme/analyzer' % self.analyzer) def _validate_vocabulary(self): vocabulary = self.vocabulary if vocabulary is not None: if isinstance(vocabulary, set): vocabulary = sorted(vocabulary) if not isinstance(vocabulary, Mapping): vocab = {} for i, t in enumerate(vocabulary): if vocab.setdefault(t, i) != i: msg = "Duplicate term in vocabulary: %r" % t raise ValueError(msg) vocabulary = vocab else: indices = set(six.itervalues(vocabulary)) if len(indices) != len(vocabulary): raise ValueError("Vocabulary contains repeated indices.") for i in xrange(len(vocabulary)): if i not in indices: msg = ("Vocabulary of size %d doesn't contain index " "%d." % (len(vocabulary), i)) raise ValueError(msg) if not vocabulary: raise ValueError("empty vocabulary passed to fit") self.fixed_vocabulary_ = True self.vocabulary_ = dict(vocabulary) else: self.fixed_vocabulary_ = False def _check_vocabulary(self): """Check if vocabulary is empty or missing (not fit-ed)""" msg = "%(name)s - Vocabulary wasn't fitted." check_is_fitted(self, 'vocabulary_', msg=msg), if len(self.vocabulary_) == 0: raise ValueError("Vocabulary is empty") class HashingVectorizer(BaseEstimator, VectorizerMixin, TransformerMixin): """Convert a collection of text documents to a matrix of token occurrences It turns a collection of text documents into a scipy.sparse matrix holding token occurrence counts (or binary occurrence information), possibly normalized as token frequencies if norm='l1' or projected on the euclidean unit sphere if norm='l2'. This text vectorizer implementation uses the hashing trick to find the token string name to feature integer index mapping. This strategy has several advantages: - it is very low memory scalable to large datasets as there is no need to store a vocabulary dictionary in memory - it is fast to pickle and un-pickle as it holds no state besides the constructor parameters - it can be used in a streaming (partial fit) or parallel pipeline as there is no state computed during fit. There are also a couple of cons (vs using a CountVectorizer with an in-memory vocabulary): - there is no way to compute the inverse transform (from feature indices to string feature names) which can be a problem when trying to introspect which features are most important to a model. - there can be collisions: distinct tokens can be mapped to the same feature index. However in practice this is rarely an issue if n_features is large enough (e.g. 2 ** 18 for text classification problems). - no IDF weighting as this would render the transformer stateful. The hash function employed is the signed 32-bit version of Murmurhash3. Read more in the :ref:`User Guide `. Parameters ---------- input : string {'filename', 'file', 'content'} If 'filename', the sequence passed as an argument to fit is expected to be a list of filenames that need reading to fetch the raw content to analyze. If 'file', the sequence items must have a 'read' method (file-like object) that is called to fetch the bytes in memory. Otherwise the input is expected to be the sequence strings or bytes items are expected to be analyzed directly. encoding : string, default='utf-8' If bytes or files are given to analyze, this encoding is used to decode. decode_error : {'strict', 'ignore', 'replace'} Instruction on what to do if a byte sequence is given to analyze that contains characters not of the given `encoding`. By default, it is 'strict', meaning that a UnicodeDecodeError will be raised. Other values are 'ignore' and 'replace'. strip_accents : {'ascii', 'unicode', None} Remove accents during the preprocessing step. 'ascii' is a fast method that only works on characters that have an direct ASCII mapping. 'unicode' is a slightly slower method that works on any characters. None (default) does nothing. analyzer : string, {'word', 'char', 'char_wb'} or callable Whether the feature should be made of word or character n-grams. Option 'char_wb' creates character n-grams only from text inside word boundaries; n-grams at the edges of words are padded with space. If a callable is passed it is used to extract the sequence of features out of the raw, unprocessed input. preprocessor : callable or None (default) Override the preprocessing (string transformation) stage while preserving the tokenizing and n-grams generation steps. tokenizer : callable or None (default) Override the string tokenization step while preserving the preprocessing and n-grams generation steps. Only applies if ``analyzer == 'word'``. ngram_range : tuple (min_n, max_n), default=(1, 1) The lower and upper boundary of the range of n-values for different n-grams to be extracted. All values of n such that min_n <= n <= max_n will be used. stop_words : string {'english'}, list, or None (default) If 'english', a built-in stop word list for English is used. If a list, that list is assumed to contain stop words, all of which will be removed from the resulting tokens. Only applies if ``analyzer == 'word'``. lowercase : boolean, default=True Convert all characters to lowercase before tokenizing. token_pattern : string Regular expression denoting what constitutes a "token", only used if ``analyzer == 'word'``. The default regexp selects tokens of 2 or more alphanumeric characters (punctuation is completely ignored and always treated as a token separator). n_features : integer, default=(2 ** 20) The number of features (columns) in the output matrices. Small numbers of features are likely to cause hash collisions, but large numbers will cause larger coefficient dimensions in linear learners. norm : 'l1', 'l2' or None, optional Norm used to normalize term vectors. None for no normalization. binary : boolean, default=False. If True, all non zero counts are set to 1. This is useful for discrete probabilistic models that model binary events rather than integer counts. dtype : type, optional Type of the matrix returned by fit_transform() or transform(). alternate_sign : boolean, optional, default True When True, an alternating sign is added to the features as to approximately conserve the inner product in the hashed space even for small n_features. This approach is similar to sparse random projection. .. versionadded:: 0.19 non_negative : boolean, optional, default False When True, an absolute value is applied to the features matrix prior to returning it. When used in conjunction with alternate_sign=True, this significantly reduces the inner product preservation property. .. deprecated:: 0.19 This option will be removed in 0.21. See also -------- CountVectorizer, TfidfVectorizer """ def __init__(self, input='content', encoding='utf-8', decode_error='strict', strip_accents=None, lowercase=True, preprocessor=None, tokenizer=None, stop_words=None, token_pattern=r"(?u)\b\w\w+\b", ngram_range=(1, 1), analyzer='word', n_features=(2 ** 20), binary=False, norm='l2', alternate_sign=True, non_negative=False, dtype=np.float64): self.input = input self.encoding = encoding self.decode_error = decode_error self.strip_accents = strip_accents self.preprocessor = preprocessor self.tokenizer = tokenizer self.analyzer = analyzer self.lowercase = lowercase self.token_pattern = token_pattern self.stop_words = stop_words self.n_features = n_features self.ngram_range = ngram_range self.binary = binary self.norm = norm self.alternate_sign = alternate_sign self.non_negative = non_negative self.dtype = dtype def partial_fit(self, X, y=None): """Does nothing: this transformer is stateless. This method is just there to mark the fact that this transformer can work in a streaming setup. """ return self def fit(self, X, y=None): """Does nothing: this transformer is stateless.""" # triggers a parameter validation if isinstance(X, six.string_types): raise ValueError( "Iterable over raw text documents expected, " "string object received.") self._get_hasher().fit(X, y=y) return self def transform(self, X): """Transform a sequence of documents to a document-term matrix. Parameters ---------- X : iterable over raw text documents, length = n_samples Samples. Each sample must be a text document (either bytes or unicode strings, file name or file object depending on the constructor argument) which will be tokenized and hashed. Returns ------- X : scipy.sparse matrix, shape = (n_samples, self.n_features) Document-term matrix. """ if isinstance(X, six.string_types): raise ValueError( "Iterable over raw text documents expected, " "string object received.") analyzer = self.build_analyzer() X = self._get_hasher().transform(analyzer(doc) for doc in X) if self.binary: X.data.fill(1) if self.norm is not None: X = normalize(X, norm=self.norm, copy=False) return X def _get_hasher(self): return FeatureHasher(n_features=self.n_features, input_type='string', dtype=self.dtype, alternate_sign=self.alternate_sign, non_negative=self.non_negative) def _document_frequency(X): """Count the number of non-zero values for each feature in sparse X.""" if sp.isspmatrix_csr(X): return np.bincount(X.indices, minlength=X.shape[1]) else: return np.diff(sp.csc_matrix(X, copy=False).indptr) class CountVectorizer(BaseEstimator, VectorizerMixin): """Convert a collection of text documents to a matrix of token counts This implementation produces a sparse representation of the counts using scipy.sparse.csr_matrix. If you do not provide an a-priori dictionary and you do not use an analyzer that does some kind of feature selection then the number of features will be equal to the vocabulary size found by analyzing the data. Read more in the :ref:`User Guide `. Parameters ---------- input : string {'filename', 'file', 'content'} If 'filename', the sequence passed as an argument to fit is expected to be a list of filenames that need reading to fetch the raw content to analyze. If 'file', the sequence items must have a 'read' method (file-like object) that is called to fetch the bytes in memory. Otherwise the input is expected to be the sequence strings or bytes items are expected to be analyzed directly. encoding : string, 'utf-8' by default. If bytes or files are given to analyze, this encoding is used to decode. decode_error : {'strict', 'ignore', 'replace'} Instruction on what to do if a byte sequence is given to analyze that contains characters not of the given `encoding`. By default, it is 'strict', meaning that a UnicodeDecodeError will be raised. Other values are 'ignore' and 'replace'. strip_accents : {'ascii', 'unicode', None} Remove accents during the preprocessing step. 'ascii' is a fast method that only works on characters that have an direct ASCII mapping. 'unicode' is a slightly slower method that works on any characters. None (default) does nothing. analyzer : string, {'word', 'char', 'char_wb'} or callable Whether the feature should be made of word or character n-grams. Option 'char_wb' creates character n-grams only from text inside word boundaries; n-grams at the edges of words are padded with space. If a callable is passed it is used to extract the sequence of features out of the raw, unprocessed input. preprocessor : callable or None (default) Override the preprocessing (string transformation) stage while preserving the tokenizing and n-grams generation steps. tokenizer : callable or None (default) Override the string tokenization step while preserving the preprocessing and n-grams generation steps. Only applies if ``analyzer == 'word'``. ngram_range : tuple (min_n, max_n) The lower and upper boundary of the range of n-values for different n-grams to be extracted. All values of n such that min_n <= n <= max_n will be used. stop_words : string {'english'}, list, or None (default) If 'english', a built-in stop word list for English is used. If a list, that list is assumed to contain stop words, all of which will be removed from the resulting tokens. Only applies if ``analyzer == 'word'``. If None, no stop words will be used. max_df can be set to a value in the range [0.7, 1.0) to automatically detect and filter stop words based on intra corpus document frequency of terms. lowercase : boolean, True by default Convert all characters to lowercase before tokenizing. token_pattern : string Regular expression denoting what constitutes a "token", only used if ``analyzer == 'word'``. The default regexp select tokens of 2 or more alphanumeric characters (punctuation is completely ignored and always treated as a token separator). max_df : float in range [0.0, 1.0] or int, default=1.0 When building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold (corpus-specific stop words). If float, the parameter represents a proportion of documents, integer absolute counts. This parameter is ignored if vocabulary is not None. min_df : float in range [0.0, 1.0] or int, default=1 When building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold. This value is also called cut-off in the literature. If float, the parameter represents a proportion of documents, integer absolute counts. This parameter is ignored if vocabulary is not None. max_features : int or None, default=None If not None, build a vocabulary that only consider the top max_features ordered by term frequency across the corpus. This parameter is ignored if vocabulary is not None. vocabulary : Mapping or iterable, optional Either a Mapping (e.g., a dict) where keys are terms and values are indices in the feature matrix, or an iterable over terms. If not given, a vocabulary is determined from the input documents. Indices in the mapping should not be repeated and should not have any gap between 0 and the largest index. binary : boolean, default=False If True, all non zero counts are set to 1. This is useful for discrete probabilistic models that model binary events rather than integer counts. dtype : type, optional Type of the matrix returned by fit_transform() or transform(). Attributes ---------- vocabulary_ : dict A mapping of terms to feature indices. 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, TfidfVectorizer 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. """ def __init__(self, input='content', encoding='utf-8', decode_error='strict', strip_accents=None, lowercase=True, preprocessor=None, tokenizer=None, stop_words=None, token_pattern=r"(?u)\b\w\w+\b", ngram_range=(1, 1), analyzer='word', max_df=1.0, min_df=1, max_features=None, vocabulary=None, binary=False, dtype=np.int64): self.input = input self.encoding = encoding self.decode_error = decode_error self.strip_accents = strip_accents self.preprocessor = preprocessor self.tokenizer = tokenizer self.analyzer = analyzer self.lowercase = lowercase self.token_pattern = token_pattern self.stop_words = stop_words self.max_df = max_df self.min_df = min_df if max_df < 0 or min_df < 0: raise ValueError("negative value for max_df or min_df") self.max_features = max_features if max_features is not None: if (not isinstance(max_features, numbers.Integral) or max_features <= 0): raise ValueError( "max_features=%r, neither a positive integer nor None" % max_features) self.ngram_range = ngram_range self.vocabulary = vocabulary self.binary = binary self.dtype = dtype def _sort_features(self, X, vocabulary): """Sort features by name Returns a reordered matrix and modifies the vocabulary in place """ sorted_features = sorted(six.iteritems(vocabulary)) map_index = np.empty(len(sorted_features), dtype=np.int32) for new_val, (term, old_val) in enumerate(sorted_features): vocabulary[term] = new_val map_index[old_val] = new_val X.indices = map_index.take(X.indices, mode='clip') return X def _limit_features(self, X, vocabulary, high=None, low=None, limit=None): """Remove too rare or too common features. Prune features that are non zero in more samples than high or less documents than low, modifying the vocabulary, and restricting it to at most the limit most frequent. This does not prune samples with zero features. """ if high is None and low is None and limit is None: return X, set() # Calculate a mask based on document frequencies dfs = _document_frequency(X) tfs = np.asarray(X.sum(axis=0)).ravel() mask = np.ones(len(dfs), dtype=bool) if high is not None: mask &= dfs <= high if low is not None: mask &= dfs >= low if limit is not None and mask.sum() > limit: mask_inds = (-tfs[mask]).argsort()[:limit] new_mask = np.zeros(len(dfs), dtype=bool) new_mask[np.where(mask)[0][mask_inds]] = True mask = new_mask new_indices = np.cumsum(mask) - 1 # maps old indices to new removed_terms = set() for term, old_index in list(six.iteritems(vocabulary)): 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 def _count_vocab(self, raw_documents, fixed_vocab): """Create sparse feature matrix, and vocabulary where fixed_vocab=False """ if fixed_vocab: vocabulary = self.vocabulary_ else: # Add a new value when a new vocabulary item is seen vocabulary = defaultdict() vocabulary.default_factory = vocabulary.__len__ analyze = self.build_analyzer() j_indices = [] indptr = _make_int_array() values = _make_int_array() indptr.append(0) for doc in raw_documents: feature_counter = {} for feature in analyze(doc): try: feature_idx = vocabulary[feature] if feature_idx not in feature_counter: feature_counter[feature_idx] = 1 else: feature_counter[feature_idx] += 1 except KeyError: # Ignore out-of-vocabulary items for fixed_vocab=True continue j_indices.extend(feature_counter.keys()) values.extend(feature_counter.values()) indptr.append(len(j_indices)) if not fixed_vocab: # disable defaultdict behaviour vocabulary = dict(vocabulary) if not vocabulary: raise ValueError("empty vocabulary; perhaps the documents only" " contain stop words") j_indices = np.asarray(j_indices, dtype=np.intc) indptr = np.frombuffer(indptr, dtype=np.intc) values = np.frombuffer(values, dtype=np.intc) X = sp.csr_matrix((values, j_indices, indptr), shape=(len(indptr) - 1, len(vocabulary)), dtype=self.dtype) X.sort_indices() return vocabulary, X def fit(self, raw_documents, y=None): """Learn a vocabulary dictionary of all tokens in the raw documents. Parameters ---------- raw_documents : iterable An iterable which yields either str, unicode or file objects. Returns ------- self """ self.fit_transform(raw_documents) return self def fit_transform(self, raw_documents, y=None): """Learn the vocabulary dictionary and return term-document matrix. This is equivalent to fit followed by transform, but more efficiently implemented. Parameters ---------- raw_documents : iterable An iterable which yields either str, unicode or file objects. Returns ------- X : array, [n_samples, n_features] Document-term matrix. """ # We intentionally don't call the transform method to make # fit_transform overridable without unwanted side effects in # TfidfVectorizer. if isinstance(raw_documents, six.string_types): raise ValueError( "Iterable over raw text documents expected, " "string object received.") self._validate_vocabulary() max_df = self.max_df min_df = self.min_df max_features = self.max_features vocabulary, X = self._count_vocab(raw_documents, self.fixed_vocabulary_) if self.binary: X.data.fill(1) if not self.fixed_vocabulary_: X = self._sort_features(X, vocabulary) n_doc = X.shape[0] max_doc_count = (max_df if isinstance(max_df, numbers.Integral) else max_df * n_doc) min_doc_count = (min_df if isinstance(min_df, numbers.Integral) else min_df * n_doc) if max_doc_count < min_doc_count: raise ValueError( "max_df corresponds to < documents than min_df") X, self.stop_words_ = self._limit_features(X, vocabulary, max_doc_count, min_doc_count, max_features) self.vocabulary_ = vocabulary return X def transform(self, raw_documents): """Transform documents to document-term matrix. Extract token counts out of raw text documents using the vocabulary fitted with fit or the one provided to the constructor. Parameters ---------- raw_documents : iterable An iterable which yields either str, unicode or file objects. Returns ------- X : sparse matrix, [n_samples, n_features] Document-term matrix. """ if isinstance(raw_documents, six.string_types): raise ValueError( "Iterable over raw text documents expected, " "string object received.") if not hasattr(self, 'vocabulary_'): self._validate_vocabulary() self._check_vocabulary() # use the same matrix-building strategy as fit_transform _, X = self._count_vocab(raw_documents, fixed_vocab=True) if self.binary: X.data.fill(1) return X def inverse_transform(self, X): """Return terms per document with nonzero entries in X. Parameters ---------- X : {array, sparse matrix}, shape = [n_samples, n_features] Returns ------- X_inv : list of arrays, len = n_samples List of arrays of terms. """ self._check_vocabulary() if sp.issparse(X): # We need CSR format for fast row manipulations. X = X.tocsr() else: # We need to convert X to a matrix, so that the indexing # returns 2D objects X = np.asmatrix(X) n_samples = X.shape[0] terms = np.array(list(self.vocabulary_.keys())) indices = np.array(list(self.vocabulary_.values())) inverse_vocabulary = terms[np.argsort(indices)] return [inverse_vocabulary[X[i, :].nonzero()[1]].ravel() for i in range(n_samples)] def get_feature_names(self): """Array mapping from feature integer indices to feature name""" self._check_vocabulary() return [t for t, i in sorted(six.iteritems(self.vocabulary_), key=itemgetter(1))] def _make_int_array(): """Construct an array.array of a type suitable for scipy.sparse indices.""" return array.array(str("i")) class TfidfTransformer(BaseEstimator, TransformerMixin): """Transform a count matrix to a normalized tf or tf-idf representation Tf means term-frequency while tf-idf means term-frequency times inverse document-frequency. This is a common term weighting scheme in information retrieval, that has also found good use in document classification. The goal of using tf-idf instead of the raw frequencies of occurrence of a token in a given document is to scale down the impact of tokens that occur very frequently in a given corpus and that are hence empirically less informative than features that occur in a small fraction of the training corpus. The formula that is used to compute the tf-idf of term t is tf-idf(d, t) = tf(t) * idf(d, t), and the idf is computed as idf(d, t) = log [ n / df(d, t) ] + 1 (if ``smooth_idf=False``), where n is the total number of documents and df(d, t) is the document frequency; the document frequency is the number of documents d that contain term t. The effect of adding "1" to the idf in the equation above is that terms with zero idf, i.e., terms that occur in all documents in a training set, will not be entirely ignored. (Note that the idf formula above differs from the standard textbook notation that defines the idf as idf(d, t) = log [ n / (df(d, t) + 1) ]). If ``smooth_idf=True`` (the default), the constant "1" is added to the numerator and denominator of the idf as if an extra document was seen containing every term in the collection exactly once, which prevents zero divisions: idf(d, t) = log [ (1 + n) / (1 + df(d, t)) ] + 1. Furthermore, the formulas used to compute tf and idf depend on parameter settings that correspond to the SMART notation used in IR as follows: Tf is "n" (natural) by default, "l" (logarithmic) when ``sublinear_tf=True``. Idf is "t" when use_idf is given, "n" (none) otherwise. Normalization is "c" (cosine) when ``norm='l2'``, "n" (none) when ``norm=None``. Read more in the :ref:`User Guide `. Parameters ---------- norm : 'l1', 'l2' or None, optional Norm used to normalize term vectors. None for no normalization. use_idf : boolean, default=True Enable inverse-document-frequency reweighting. smooth_idf : boolean, default=True Smooth idf weights by adding one to document frequencies, as if an extra document was seen containing every term in the collection exactly once. Prevents zero divisions. sublinear_tf : boolean, default=False Apply sublinear tf scaling, i.e. replace tf with 1 + log(tf). References ---------- .. [Yates2011] `R. Baeza-Yates and B. Ribeiro-Neto (2011). Modern Information Retrieval. Addison Wesley, pp. 68-74.` .. [MRS2008] `C.D. Manning, P. Raghavan and H. Schütze (2008). Introduction to Information Retrieval. Cambridge University Press, pp. 118-120.` """ def __init__(self, norm='l2', use_idf=True, smooth_idf=True, sublinear_tf=False): self.norm = norm self.use_idf = use_idf self.smooth_idf = smooth_idf self.sublinear_tf = sublinear_tf def fit(self, X, y=None): """Learn the idf vector (global term weights) Parameters ---------- X : sparse matrix, [n_samples, n_features] a matrix of term/token counts """ if not sp.issparse(X): X = sp.csc_matrix(X) if self.use_idf: n_samples, n_features = X.shape df = _document_frequency(X) # perform idf smoothing if required df += int(self.smooth_idf) n_samples += int(self.smooth_idf) # log+1 instead of log makes sure terms with zero idf don't get # suppressed entirely. idf = np.log(float(n_samples) / df) + 1.0 self._idf_diag = sp.spdiags(idf, diags=0, m=n_features, n=n_features, format='csr') return self def transform(self, X, copy=True): """Transform a count matrix to a tf or tf-idf representation Parameters ---------- X : sparse matrix, [n_samples, n_features] a matrix of term/token counts copy : boolean, default True Whether to copy X and operate on the copy or perform in-place operations. Returns ------- vectors : sparse matrix, [n_samples, n_features] """ if hasattr(X, 'dtype') and np.issubdtype(X.dtype, np.floating): # preserve float family dtype X = sp.csr_matrix(X, copy=copy) else: # convert counts or binary occurrences to floats X = sp.csr_matrix(X, dtype=np.float64, copy=copy) n_samples, n_features = X.shape if self.sublinear_tf: np.log(X.data, X.data) X.data += 1 if self.use_idf: check_is_fitted(self, '_idf_diag', 'idf vector is not fitted') expected_n_features = self._idf_diag.shape[0] if n_features != expected_n_features: raise ValueError("Input has n_features=%d while the model" " has been trained with n_features=%d" % ( n_features, expected_n_features)) # *= doesn't work X = X * self._idf_diag if self.norm: X = normalize(X, norm=self.norm, copy=False) return X @property def idf_(self): # if _idf_diag is not set, this will raise an attribute error, # which means hasattr(self, "idf_") is False return np.ravel(self._idf_diag.sum(axis=0)) class TfidfVectorizer(CountVectorizer): """Convert a collection of raw documents to a matrix of TF-IDF features. Equivalent to CountVectorizer followed by TfidfTransformer. Read more in the :ref:`User Guide `. Parameters ---------- input : string {'filename', 'file', 'content'} If 'filename', the sequence passed as an argument to fit is expected to be a list of filenames that need reading to fetch the raw content to analyze. If 'file', the sequence items must have a 'read' method (file-like object) that is called to fetch the bytes in memory. Otherwise the input is expected to be the sequence strings or bytes items are expected to be analyzed directly. encoding : string, 'utf-8' by default. If bytes or files are given to analyze, this encoding is used to decode. decode_error : {'strict', 'ignore', 'replace'} Instruction on what to do if a byte sequence is given to analyze that contains characters not of the given `encoding`. By default, it is 'strict', meaning that a UnicodeDecodeError will be raised. Other values are 'ignore' and 'replace'. strip_accents : {'ascii', 'unicode', None} Remove accents during the preprocessing step. 'ascii' is a fast method that only works on characters that have an direct ASCII mapping. 'unicode' is a slightly slower method that works on any characters. None (default) does nothing. analyzer : string, {'word', 'char'} or callable Whether the feature should be made of word or character n-grams. If a callable is passed it is used to extract the sequence of features out of the raw, unprocessed input. preprocessor : callable or None (default) Override the preprocessing (string transformation) stage while preserving the tokenizing and n-grams generation steps. tokenizer : callable or None (default) Override the string tokenization step while preserving the preprocessing and n-grams generation steps. Only applies if ``analyzer == 'word'``. ngram_range : tuple (min_n, max_n) The lower and upper boundary of the range of n-values for different n-grams to be extracted. All values of n such that min_n <= n <= max_n will be used. stop_words : string {'english'}, list, or None (default) If a string, it is passed to _check_stop_list and the appropriate stop list is returned. 'english' is currently the only supported string value. If a list, that list is assumed to contain stop words, all of which will be removed from the resulting tokens. Only applies if ``analyzer == 'word'``. If None, no stop words will be used. max_df can be set to a value in the range [0.7, 1.0) to automatically detect and filter stop words based on intra corpus document frequency of terms. lowercase : boolean, default True Convert all characters to lowercase before tokenizing. token_pattern : string Regular expression denoting what constitutes a "token", only used if ``analyzer == 'word'``. The default regexp selects tokens of 2 or more alphanumeric characters (punctuation is completely ignored and always treated as a token separator). max_df : float in range [0.0, 1.0] or int, default=1.0 When building the vocabulary ignore terms that have a document frequency strictly higher than the given threshold (corpus-specific stop words). If float, the parameter represents a proportion of documents, integer absolute counts. This parameter is ignored if vocabulary is not None. min_df : float in range [0.0, 1.0] or int, default=1 When building the vocabulary ignore terms that have a document frequency strictly lower than the given threshold. This value is also called cut-off in the literature. If float, the parameter represents a proportion of documents, integer absolute counts. This parameter is ignored if vocabulary is not None. max_features : int or None, default=None If not None, build a vocabulary that only consider the top max_features ordered by term frequency across the corpus. This parameter is ignored if vocabulary is not None. vocabulary : Mapping or iterable, optional Either a Mapping (e.g., a dict) where keys are terms and values are indices in the feature matrix, or an iterable over terms. If not given, a vocabulary is determined from the input documents. binary : boolean, default=False If True, all non-zero term counts are set to 1. This does not mean outputs will have only 0/1 values, only that the tf term in tf-idf is binary. (Set idf and normalization to False to get 0/1 outputs.) dtype : type, optional Type of the matrix returned by fit_transform() or transform(). norm : 'l1', 'l2' or None, optional Norm used to normalize term vectors. None for no normalization. use_idf : boolean, default=True Enable inverse-document-frequency reweighting. smooth_idf : boolean, default=True Smooth idf weights by adding one to document frequencies, as if an extra document was seen containing every term in the collection exactly once. Prevents zero divisions. sublinear_tf : boolean, default=False Apply sublinear tf scaling, i.e. replace tf with 1 + log(tf). Attributes ---------- vocabulary_ : dict A mapping of terms to feature indices. idf_ : array, shape = [n_features], or None The learned idf vector (global term weights) when ``use_idf`` is set to True, None otherwise. 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 Tokenize the documents and count the occurrences of token and return them as a sparse matrix TfidfTransformer Apply Term Frequency Inverse Document Frequency normalization to a sparse matrix of occurrence 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. """ def __init__(self, input='content', encoding='utf-8', decode_error='strict', strip_accents=None, lowercase=True, preprocessor=None, tokenizer=None, analyzer='word', stop_words=None, token_pattern=r"(?u)\b\w\w+\b", ngram_range=(1, 1), max_df=1.0, min_df=1, max_features=None, vocabulary=None, binary=False, dtype=np.int64, norm='l2', use_idf=True, smooth_idf=True, sublinear_tf=False): super(TfidfVectorizer, self).__init__( input=input, encoding=encoding, decode_error=decode_error, strip_accents=strip_accents, lowercase=lowercase, preprocessor=preprocessor, tokenizer=tokenizer, analyzer=analyzer, stop_words=stop_words, token_pattern=token_pattern, ngram_range=ngram_range, max_df=max_df, min_df=min_df, max_features=max_features, vocabulary=vocabulary, binary=binary, dtype=dtype) self._tfidf = TfidfTransformer(norm=norm, use_idf=use_idf, smooth_idf=smooth_idf, sublinear_tf=sublinear_tf) # Broadcast the TF-IDF parameters to the underlying transformer instance # for easy grid search and repr @property def norm(self): return self._tfidf.norm @norm.setter def norm(self, value): self._tfidf.norm = value @property def use_idf(self): return self._tfidf.use_idf @use_idf.setter def use_idf(self, value): self._tfidf.use_idf = value @property def smooth_idf(self): return self._tfidf.smooth_idf @smooth_idf.setter def smooth_idf(self, value): self._tfidf.smooth_idf = value @property def sublinear_tf(self): return self._tfidf.sublinear_tf @sublinear_tf.setter def sublinear_tf(self, value): self._tfidf.sublinear_tf = value @property def idf_(self): return self._tfidf.idf_ def fit(self, raw_documents, y=None): """Learn vocabulary and idf from training set. Parameters ---------- raw_documents : iterable an iterable which yields either str, unicode or file objects Returns ------- self : TfidfVectorizer """ X = super(TfidfVectorizer, self).fit_transform(raw_documents) self._tfidf.fit(X) return self def fit_transform(self, raw_documents, y=None): """Learn vocabulary and idf, return term-document matrix. This is equivalent to fit followed by transform, but more efficiently implemented. Parameters ---------- raw_documents : iterable an iterable which yields either str, unicode or file objects Returns ------- X : sparse matrix, [n_samples, n_features] Tf-idf-weighted document-term matrix. """ X = super(TfidfVectorizer, self).fit_transform(raw_documents) self._tfidf.fit(X) # X is already a transformed view of raw_documents so # we set copy to False return self._tfidf.transform(X, copy=False) def transform(self, raw_documents, copy=True): """Transform documents to document-term matrix. Uses the vocabulary and document frequencies (df) learned by fit (or fit_transform). Parameters ---------- raw_documents : iterable an iterable which yields either str, unicode or file objects copy : boolean, default True Whether to copy X and operate on the copy or perform in-place operations. Returns ------- X : sparse matrix, [n_samples, n_features] Tf-idf-weighted document-term matrix. """ check_is_fitted(self, '_tfidf', 'The tfidf vector is not fitted') X = super(TfidfVectorizer, self).transform(raw_documents) return self._tfidf.transform(X, copy=False)