def read_scores(all=False, remove_null=False, data_dir=RAW_DATA_DIR):
file_path = os.path.join(data_dir,
'lqual{}.jsonl'.format('_all' if all else ''))
if not os.path.isfile(file_path):
raise ValueError("The file {} does not exist!".format(file_path))
scores = []
id_set = set()
with jsonl.open(file_path) as file:
for sample in file:
entry = OrderedDict()
entry['id'] = sample['input']['contents']['id']
entry['ref'] = sample['input']['contents']['reference']
entry['sys_name'] = sample['input']['contents']['system']
entry['sys_summ'] = sample['input']['contents']['text']
del sample['output']['_responses']
entry['scores'] = sample['output']
if remove_null and entry['scores']['overall'] is None:
continue
scores.append(entry)
id_set.add(entry['id'])
return scores, list(id_set)
def read_article_refs(ids=None, as_dict=False, data_dir=RAW_DATA_DIR):
refs, ref_ids = read_references()
article_path = os.path.join(data_dir, 'articles.jsonl')
if not os.path.isfile(article_path):
raise ValueError("The file {} does not exist!".format(article_path))
with jsonl.open(article_path) as file:
for sample in file:
if sample['id'] not in ref_ids:
continue
idx = find_idx_by_id(refs, sample['id'])
assert idx != -1
refs[idx]['article'] = sample['text']
if ids is not None:
refs = [r for r in refs if r['id'] in ids]
if as_dict:
article_ref_dict = {}
for r in refs:
article_ref_dict.update(
{r['id']: {
'article': r['article'],
'ref': r['ref']
}})
return article_ref_dict
else:
return refs
def read_samples(data_dir=PROCESSED_DATA_DIR):
sample_file_path = os.path.join(data_dir, 'samples.jsonl.gz')
if not os.path.isfile(sample_file_path):
raise ValueError(
"The file {} does not exist!".format(sample_file_path))
with jsonl.open(sample_file_path, gzip=True) as sample_file:
return sample_file.read()
def read_articles(data_dir=RAW_DATA_DIR):
article_path = os.path.join(data_dir, 'articles.jsonl')
if not os.path.isfile(article_path):
raise ValueError("The file {} does not exist!".format(article_path))
articles = []
article_id_list = []
with jsonl.open(article_path) as article_file:
for sample in article_file:
entry = OrderedDict()
entry['id'] = sample['id']
entry['article'] = sample['text']
articles.append(entry)
if entry['id'] not in article_id_list:
article_id_list.append(entry['id'])
return articles, article_id_list
def read_processed_scores(all=False,
remove_null=False,
data_dir=PROCESSED_DATA_DIR):
file_path = os.path.join(data_dir,
'lqual{}.jsonl'.format('_all' if all else ''))
if not os.path.isfile(file_path):
raise ValueError("The file {} does not exist!".format(file_path))
scores, _ = read_scores(all, remove_null)
scores_dict = {}
for score in scores:
key = "{}-{}".format(score['id'], score['sys_name'])
if key not in scores_dict:
scores_dict.update({key: score})
with jsonl.open(file_path) as file:
for sample in file:
key = "{}-{}".format(sample['id'], sample['system'])
if key in scores_dict:
del sample['prompts']['overall']['gold']
del sample['prompts']['overall']['human']
scores_dict[key]['metrics'] = sample['prompts']['overall']
del_keys = []
for key, entry in scores_dict.items():
if 'metrics' not in entry:
del_keys.append(key)
for key in del_keys:
scores_dict.pop(key)
id_list = list(set([entry['id'] for entry in scores_dict.values()]))
return list(scores_dict.values()), id_list
def read_references(data_dir=RAW_DATA_DIR):
ref_path = os.path.join(data_dir, 'lqual_all.jsonl')
if not os.path.isfile(ref_path):
raise ValueError("The file {} does not exist!".format(ref_path))
refs = []
id_list = []
with jsonl.open(ref_path) as file:
for sample in file:
if sample['id'] in id_list:
continue
entry = OrderedDict()
entry['id'] = sample['id']
entry['ref'] = sample['reference']
refs.append(entry)
id_list.append(entry['id'])
return refs, id_list
