def observed_norm(self, observed):
"""Accepts a sparse vector of word, list of (word int,count) 2-tuples.
Returns real number between -2 and 2.
"""
numpos = 0
numneg = 0
for n,word,count in topiclib.iterwords(observed):
if word in posi:
numpos += 1
if word in negi:
numneg += 1
ratio = 1.0
normratio = 0.0
if numpos == 0 and numneg == 0:
return 0.0
elif numneg == 0:
return 2.0
elif numpos == 0:
return -2.0
else:
if numpos >= numneg:
ratio = float(numpos) / numneg
normratio = (ratio - 1)
else:
ratio = -1.0 * float(numneg) / numpos
normratio = (ratio + 1)
o = graphlib.logistic_sigmoid(normratio)
return (4 * o) - 2 # norm to -2 to 2
def test_iterwords():
doc0 = [(0,3), (1,1)]
doc1 = [(1,3), (2,2),(0,1)]
out = list(lm.iterwords(doc0))
answer = [
(0, 0, 3),
(1, 0, 3),
(2, 0, 3),
(3, 1, 1),
]
assert out == answer
out2 = list(lm.iterwords(doc1))
answer2 = [
(0, 1, 3),
(1, 1, 3),
(2, 1, 3),
(3, 2, 2),
(4, 2, 2),
(5, 0, 1),
]
assert out2 == answer2
