def main(args):
word_embs = word_embeddings.load_embeddings(args.embs_path)
system_summaries = []
references_list = []
with open(args.input_jsonl, 'r') as f:
for line in f:
instance = json.loads(line)
# The input summaries to S3 are lists of sentences. The example
# just passes the whole text in as 1 sentence without pre-sentence tokenizing
# it, so we will do the same. But the input summaries are expected
# to just be 1 string each, so we wrap them in an extra list
summary = [instance['summary']]
references = [[reference] for reference in instance['references']]
system_summaries.append(summary)
references_list.append(references)
scores_pyr, scores_resp = S3.S3_batch(references_list, system_summaries,
word_embs, args.model_folder)
with open(args.output_jsonl, 'w') as out:
for pyr, resp in zip(scores_pyr, scores_resp):
out.write(json.dumps({'pyr': pyr, 'resp': resp}) + '\n')
