def _markers(self, window=3):
text = self.text().decode('utf-8')
pars = paragraphs(text)
df = pd.DataFrame({
'bio_word_match': matches(pars, self.BIO_WORDS),
'text': pars,
})
last_names = self._director_last_names()
for k in last_names:
assert k not in df
density = lambda s: float(len(k)) * float(s.lower().count(k.lower())) / float(len(s))
df[k] = df.text.map(density)
df['one_name'] = df[last_names].apply(max, 1)
# Mark 5 paragraphs down from where we see a last name match
df['name'] = pd.rolling_max(df.one_name, window=window, center=True)
# Mark 5 paragraphs around where we see a bio word
df['bio'] = pd.rolling_max(df.bio_word_match, window=window, center=True)
# Multiply them together and see what we have
df['mark'] = pd.rolling_mean(df.name * df.bio, window=window, center=True)
s = np.zeros(len(df.mark))
for i, flag in enumerate(df.mark):
if flag:
if s[i-1]:
s[i] = s[i-1]
else:
s[i] = i
df['group'] = s
s = df.group.value_counts()
g = s[s.index != 0].argmax()
df['flag'] = df.group == g
return df
def matching_paragraphs(text, last_names):
pattern = '(%s)' % '|'.join(last_names)
matching = [p for p in paragraphs(text) if re.search(pattern, p, re.IGNORECASE)]
return matching
