def test_iter(self):
#Calculate original
selector_original = SelectKBest(f_classif, k=1)
selector_original.fit(self.X, self.y)
#Calculate custom
X_y = izip(self.X, self.y)
selector = iSelectKBest(if_classif, k=1)
selector.fit(X_y, self.X.shape[1])
#Asserts
np.testing.assert_array_almost_equal([0.05882353, 0.03846154, 0.17241379],
selector.scores_, 8)
np.testing.assert_array_almost_equal([0.83096915, 0.86263944, 0.71828192],
selector.pvalues_, 8)
#Asserts
np.testing.assert_array_equal(selector_original.scores_,
selector.scores_)
np.testing.assert_array_equal(selector_original.pvalues_,
selector.pvalues_)
def __init__(self, corpus, document_titles,
test_corpus, test_corpus_targets, num_test_corpus,
num_best_features = 1000,
num_features = None,
tmp_path = 'complete_similarity'):
"""
The similarity between a document and each document of the corpus is
the feature created.
The corpus is filtered for significant features.
Parameters
----------
corpus : The corpus which contains all concepts. E.g. English Wikipedia
in TF-IDF space.
test_corpus : The test corpus is used to select features.
All documents in this corous should be classified.
test_corpus_targets : The target classes of each document in the
test corpus.
num_test_corpus : Number of documents in the test corpus.
document_titles : The names of each concept (doc) in corpus.
num_features : The number of features of corpus
num_best_features : Number of features which should be selected for cESA model.
If one wants to use all concepts as features the she has
to set num_best_features to the size of corpus.
"""
if num_features is None:
logger.info("scanning corpus to determine the number of features")
num_features = 1 + utils.get_max_id(corpus)
self.num_features = num_features
#create similarity index of complete corpus
complete_similarity_index = Similarity(output_prefix = tmp_path,
corpus = corpus,
num_features = self.num_features)
#reduce concept count by feature selection
self.selector = iSelectKBest(if_classif, k = num_best_features)
#transform each document of test_corpus
logger.info("Test corpus of %d documents..." % num_test_corpus)
transformed_test_corpus = (complete_similarity_index[doc]
for doc
in test_corpus)
logger.info("Select best features...")
X_y = izip(transformed_test_corpus, test_corpus_targets)
self.selector.fit(X_y, len(document_titles))
logger.info("Done selecting.")
#reduce similarity index
selected_documents = [doc
for doc, mask
in itertools.izip(corpus, self.selector.get_support())
if mask]
self.similarity_index = MatrixSimilarity(corpus = selected_documents,
num_features = self.num_features)
#reduce document titles
self.document_titles = DocumentTitles()
for doc_title, mask in itertools.izip(document_titles, self.selector.get_support()):
if mask:
self.document_titles.append(doc_title)
#print doc titles
for title in self.document_titles:
logger.debug("%s" % title)