Add lower bound wilson score for bernoulli parameter

[c-martinez]

Alternative way of converting counts to probabilities (as suggested by #13 ).

In the following example, entities like: Chiliometrum, Magnum opus, Litus, Occidens, Septentrio, etc, are assigned 1.0 probability when using 'simple' count conversion, but get different probabilities when using 'wilson' method.

import re
from semanticizest import Semanticizer

sem = Semanticizer('la.model', score='wilson', wilson_confidence=0.95)

text = """Area 389.434 km² Naxos est maxima Cycladum insula. Insulae orientali sunt litora ardua, in occidentem versus loca planiora patent, a septentrionibus ad meridiem montes granitici insulam transeunt, qui usque ad 1000 metra surgunt; quorum summa cacumina sunt Mons Iovis et Coronus."""
toks = re.findall('\w+', text)

for cand in sem.all_candidates(toks):
   print cand

However, it is significantly slower.

[larsmans]

Should add scipy to requirements.txt.