def get_datum_in_doc(self, doc, i):
# Read the doc for i, if it's not already in memory
# Get it out of the doc, return it as a column vector
if self.current_doc != doc:
self.current_doc = doc
self.current_base_index = self.doc_index[doc, 0]
self.doc_data = self.read_doc(doc)
return asvector(self.doc_data[:, i-self.current_base_index])
def corpus_mean(reader):
sum = np.zeros((reader.dim,), 'float64')
for d in range(reader.num_docs):
data = reader.read_doc(d)
np.add(sum, data.sum(axis=1), out=sum)
return asvector(sum / reader.num_data)