def scorer(lucene_score, query_id) -> NamedNumber:
claim_id, p_id = query_id.split("_")
p_text = perspective_getter(int(p_id))
tokens = tokenizer.tokenize_stem(p_text)
c_lm = claim_log_odds_dict[claim_id]
reason = " ".join(["{0} ({1:.2f})".format(t, c_lm[t]) for t in tokens])
score = sum([c_lm[t] for t in tokens])
return NamedNumber(score, reason)
def scorer(lucene_score, query_id) -> NamedNumber:
claim_id, p_id = query_id.split("_")
gold_pids = gold[int(claim_id)]
score = 0
for p_ids in gold_pids:
if int(p_id) in p_ids:
score = 1
return NamedNumber(score, "")
def scorer(query_p: Passage,
candidate: List[Passage]) -> List[NamedNumber]:
payload = []
text1 = query_p.text.splitlines()[0]
print(text1)
for c in candidate:
payload.append((text1, c.text))
r = client.request_multiple(payload)
r = [NamedNumber(v, "") for v in r]
return r
def scorer(lucene_score, query_id) -> NamedNumber:
claim_id, p_id = query_id.split("_")
i_claim_id = int(claim_id)
payload = []
p_text = perspective_getter(int(p_id))
c_text = cid_to_text[i_claim_id]
payload.append(encoder.encode_pair(c_text, p_text))
r = proxy.predict(payload)
ns_score = -float(r[0])
#ns_score = 0
score = bm25_module.score(c_text, p_text)
new_score = score + ns_score * 10
score = NamedNumber(new_score, score.name + " {}".format(ns_score))
return score
def scorer(lucene_score, query_id) -> NamedNumber:
claim_id, p_id = query_id.split("_")
c_text = cid_to_text[int(claim_id)]
p_text = perspective_getter(int(p_id))
score: NamedNumber = bm25_module.score(c_text, p_text)
nclaim_id = int(claim_id)
if nclaim_id in rm_info:
ex_qtf = rm_info_c[nclaim_id]
p_tokens = tokenizer.tokenize_stem(p_text)
ex_score = bm25_module.score_inner(ex_qtf, Counter(p_tokens))
new_info = score.name + "({})".format(ex_score.name)
score = NamedNumber(score + ex_score, new_info)
else:
not_found.add(claim_id)
return score
def score_inner(self, q_tf, t_tf) -> NamedNumber:
dl = sum(t_tf.values())
score_sum = 0
info = []
for q_term, qtf in q_tf.items():
t = BM25_verbose(f=t_tf[q_term],
qf=qtf,
df=self.df[q_term],
N=self.N,
dl=dl,
avdl=self.avdl,
b=self.b,
my_k1=self.k1,
my_k2=self.k2)
score_sum += t
info.append((q_term, t))
ideal_score = 0
for q_term, qtf in q_tf.items():
max_t = BM25_verbose(f=qtf,
qf=qtf,
df=self.df[q_term],
N=self.N,
dl=dl,
avdl=self.avdl,
b=self.b,
my_k1=self.k1,
my_k2=self.k2)
ideal_score += max_t
info_log = "Ideal Score={0:.1f} ".format(ideal_score)
info.sort(key=lambda x: x[1], reverse=True)
for q_term, t in info:
if t > 0.001:
info_log += "{0}({1:.2f}) ".format(q_term, t)
return NamedNumber(score_sum, info_log)
def do_score(candidate_p: Passage) -> NamedNumber:
if candidate_p.text == query_p.text:
return NamedNumber(-99, "equal")
p_tf = basic_tf.get_tf(candidate_p)
return bm25_module.score_inner(q_tf, p_tf)
def scorer(claim_id: int, p_tokens: List[str]) -> NamedNumber:
c_lm = claim_log_odds_dict[claim_id]
reason = " ".join(["{0} ({1:.2f})".format(t, c_lm[t]) for t in p_tokens])
score = sum([c_lm[t] for t in p_tokens])
return NamedNumber(score, reason)