import gzip
import random
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
random.seed(42)

for dataset in 'dev-0', 'test-A':
    with gzip.open(f'{dataset}/in.tsv.gz', 'rt') as f_in, open(f'{dataset}/out.tsv', 'w') as f_out:
        for line in f_in:
            line = line.rstrip('\n').split('\t')
            query = line[0]
            candidates = line[1:]

            vectorizer = TfidfVectorizer()
            query_v = vectorizer.fit_transform([query.replace('[SEP]', ' ')])
            candidates_v = vectorizer.transform(candidates)
            similarites = cosine_similarity(query_v, candidates_v)[0]
            candidates_sorted = sorted(zip(candidates, similarites), key= lambda x: x[1], reverse=True)
            candidates_sorted = [c[0] for c in candidates_sorted]

            candidates = '\t'.join(candidates_sorted) + '\n'
            f_out.write(candidates)