def percentage_of_concept_ans_in_doc(self, include_extended=False):
"""
Where answer is found in the relevant passage.
"""
from baselines import read_concepts
n_all = 0
n_found = 0
for datum in self.dataset[DATA_KEY]:
title_and_passage = datum[DOC_KEY][TITLE_KEY] + "\n" + datum[
DOC_KEY][CONTEXT_KEY]
concept_set = {
" ".join(concept).lower()
for concept in read_concepts(title_and_passage)
}
for qa in datum[DOC_KEY][QAS_KEY]:
ans = qa[ANS_KEY]
a = ""
for _a in ans:
if _a[ORIG_KEY] == "dataset":
a = _a[TXT_KEY]
assert a
if a.lower() in concept_set:
n_found += 1
elif include_extended:
is_found = False
for _a in ans:
if _a[ORIG_KEY] == "UMLS":
a = _a[TXT_KEY]
if a.lower() in concept_set:
is_found = True
if is_found:
n_found += 1
n_all += 1
return 100 * n_found / n_all
def entities_passage(self, lowercase=True):
from baselines import read_concepts
e = Counter()
for datum in self.dataset[DATA_KEY]:
title_and_passage = datum[DOC_KEY][TITLE_KEY] + "\n" + datum[
DOC_KEY][CONTEXT_KEY]
e.update([
to_lower(" ".join(concept), lowercase)
for concept in read_concepts(title_and_passage)
])
return e
def entities(self, lowercase=True, include_extended=False):
from baselines import read_concepts
e = Counter()
for datum in self.dataset[DATA_KEY]:
title_and_passage = datum[DOC_KEY][TITLE_KEY] + "\n" + datum[
DOC_KEY][CONTEXT_KEY]
e.update([
to_lower(" ".join(concept), lowercase)
for concept in read_concepts(title_and_passage)
])
for datum in self.dataset[DATA_KEY]:
for qa in datum[DOC_KEY][QAS_KEY]:
e.update([
to_lower(" ".join(concept), lowercase)
for concept in read_concepts(qa[QUERY_KEY])
])
for a in qa[ANS_KEY]:
if a[ORIG_KEY] == "dataset" or include_extended:
e[to_lower(a[TXT_KEY], lowercase)] += 1
return e