corpus = "/path/to/text/corpus" #note: positive and negative docs are in seperate folders
label_names = ['positive', 'negative']

textfiles = load_files(corpus, categories=label_names, load_content=False)
data = [cs.open(filename, 'r', 'UTF-8').read() for filename in notices.filenames]

#Tokenize and get counts for all documents, becomes a numpy data structure
#in a format suitable for scikit-learn
countVec = CountVectorizer(stop_words="english", binary=True,
                           ngram_range=(1, 1), strip_accents='unicode')
X = countVec.fit_transform(data)
Y = texfiles.target       

#the counts are now used to compute the BNS score, we are ready to classify/cluster/whatever
vocab = countVec.vocabulary_
bns = BnsTransformer(y=Y, vocab=vocab)
X_bns = bns.transform(X)