def gen_freqstats(argv):
""" Generate frequency stats """
parser = argparse.ArgumentParser(
prog='gen_freqstats',
formatter_class=argparse.RawDescriptionHelpFormatter,
add_help=False,
)
parser.add_argument('-m', dest='model', metavar='DIR', required=True,
help='store the processed data in DIR')
parser.add_argument('index_path',
help='path to Galago index')
parser.add_argument('index_part',
help='index part: postings.krovetz or postings.porter')
args = parser.parse_args(argv)
model = summaryrank.Model(args.model)
term_set = set()
for text, _ in model.load_topics('topics_stem'):
term_set.update(text.split())
for text, _ in model.load_sentences('sentences_stem'):
term_set.update(text.split())
print >>sys.stderr, 'found {} stems'.format(len(term_set))
GalagoIndexDump.dump(model.get_path('freq_stats'), args.index_path, args.index_part, term_set)
def compute(self, model):
result = []
if not self._freq_stats:
self._freq_stats = GalagoIndexDump.load(model.get_path('freq_stats'))
collection_len = self._freq_stats.collection_length()
topics_stem = model.load_topics('topics_stem')
queries = dict((m['qid'], text.split()) for text, m in topics_stem)
sentences_stem = model.load_sentences('sentences_stem')
for text, m in sentences_stem:
stems = text.split()
sentence_tf = collections.Counter(stems)
sentence_len = len(stems)
score = float(0)
for query_stem in queries[m['qid']]:
cf = self._freq_stats.cf(query_stem)
if cf == 0:
continue
score += math.log(
float(sentence_tf[query_stem] + self.mu * float(cf) / collection_len)
/ (sentence_len + self.mu))
result.append(score)
return result
def compute(self, model):
result = []
if not self._freq_stats:
self._freq_stats = GalagoIndexDump.load(model.get_path('freq_stats'))
N = self._freq_stats.num_docs()
topics_stem = model.load_topics('topics_stem')
queries = dict((m['qid'], text.split()) for text, m in topics_stem)
for text, m in model.load_sentences('sentences_stem'):
stems = text.split()
sentence_tf = collections.Counter(stems)
sentence_len = len(stems)
score = float(0)
for query_stem in queries[m['qid']]:
df = self._freq_stats.df(query_stem)
comp1 = math.log(float(N - df + 0.5) / (df + 0.5))
comp2 = float(sentence_tf[query_stem] * (self.k1 + 1))
comp3 = sentence_tf[query_stem] + \
self.k1 * (1 - self.b + float(self.b * sentence_len) / self.avgdl)
score += comp1 * comp2 / comp3
result.append(score)
return result