def set(self, article, context, candidate):
"""
set feature attribute to:
- greedy score (dim: 1) between candidate response & last k user turns without stop words
- average embedding score (dim: 1) between candidate response & last k user turns without stop words
- extrema embedding score (dim: 1) between candidate response & last k user turns without stop words
"""
if candidate is None or context is None:
self.feat = None
else:
candidate = candidate.lower()
start = min(len(context), 2 * self.k + 1)
user_turns = np.array(context)[range(-start, 0, 2)]
user_turns = ' '.join(user_turns).lower()
user_turns = ' '.join(
filter(lambda word: word not in stop,
word_tokenize(user_turns)))
logger.debug("last %d user turns: %s" % (self.k, user_turns))
res1 = greedy_score(candidate, user_turns)
res2 = greedy_score(user_turns, candidate)
self.feat = np.zeros(3)
self.feat[0] = (res1 + res2) / 2.0
self.feat[1] = float(average_score(candidate, user_turns))
self.feat[2] = float(extrema_score(candidate, user_turns))
def set(self, article, context, candidate):
"""
set feature attribute to:
- greedy score (dim: 1) between candidate response & article
- average embedding score (dim: 1) between candidate response & article
- extrema embedding score (dim: 1) between candidate response & article
"""
if candidate is None or article is None:
self.feat = None
else:
candidate = candidate.lower()
article = article.lower()
res1 = greedy_score(candidate, article)
res2 = greedy_score(article, candidate)
self.feat = np.zeros(3)
self.feat[0] = (res1 + res2) / 2.0
self.feat[1] = float(average_score(candidate, article))
self.feat[2] = float(extrema_score(candidate, article))
def set(self, article, context, candidate):
"""
set feature attribute to:
- greedy score (dim: 1) between candidate response & last user turn
- average embedding score (dim: 1) between candidate response & last user turn
- extrema embedding score (dim: 1) between candidate response & last user turn
"""
if candidate is None or context is None:
self.feat = None
else:
candidate = candidate.lower()
last_turn = context[-1].lower()
res1 = greedy_score(candidate, last_turn)
res2 = greedy_score(last_turn, candidate)
self.feat = np.zeros(3)
self.feat[0] = (res1 + res2) / 2.0
self.feat[1] = float(average_score(candidate, last_turn))
self.feat[2] = float(extrema_score(candidate, last_turn))
def set(self, article, context, candidate):
"""
set feature attribute to:
- greedy score (dim: 1) between candidate response & last k turns
- average embedding score (dim: 1) between candidate response & last k turns
- extrema embedding score (dim: 1) between candidate response & last k turns
"""
if candidate is None or context is None:
self.feat = None
else:
candidate = candidate.lower()
last_turns = ' '.join(context[-self.k:]).lower()
logger.debug("last %d turns: %s" % (self.k, last_turns))
res1 = greedy_score(candidate, last_turns)
res2 = greedy_score(last_turns, candidate)
self.feat = np.zeros(3)
self.feat[0] = (res1 + res2) / 2.0
self.feat[1] = float(average_score(candidate, last_turns))
self.feat[2] = float(extrema_score(candidate, last_turns))
def set(self, article, context, candidate):
"""
set feature attribute to:
- greedy score (dim: 1) between candidate response & article without stop words
- average embedding score (dim: 1) between candidate response & article without stop words
- extrema embedding score (dim: 1) between candidate response & article without stop words
"""
if candidate is None or article is None:
self.feat = None
else:
candidate = candidate.lower()
article = article.lower()
article = ' '.join(
filter(lambda word: word not in stop, word_tokenize(article)))
res1 = greedy_score(candidate, article)
res2 = greedy_score(article, candidate)
self.feat = np.zeros(3)
self.feat[0] = (res1 + res2) / 2.0
self.feat[1] = float(average_score(candidate, article))
self.feat[2] = float(extrema_score(candidate, article))