def test_rerank_hypotheses(hypotheses, reference, expected_output, metric):
reranker = rerank.Reranker(metric=metric, return_score=False)
reranked = reranker.rerank_hypotheses(hypotheses, reference)
actual_list = reranked.hypotheses
assert actual_list == expected_output
def test_rerank_return_score(hypotheses, reference, expected_scores):
reranker = rerank.Reranker(metric="bleu", return_score=True)
hypotheses = {'translations': hypotheses}
reranked_hypotheses = reranker.rerank(hypotheses, reference)
assert 'scores' in reranked_hypotheses
actual_scores = reranked_hypotheses['scores']
assert np.allclose(actual_scores, expected_scores)
def test_rerank_return_score(hypotheses, reference, expected_scores):
reranker = rerank.Reranker(metric="bleu", return_score=True)
reranked_with_scores = reranker.rerank_hypotheses(hypotheses, reference)
actual_scores = reranked_with_scores.scores
assert np.allclose(actual_scores, expected_scores)
def test_rerank_hypotheses(hypotheses, reference, expected_output, metric):
reranker = rerank.Reranker(metric=metric, return_score=False)
hypotheses = {
'sentence_id': 0,
'translation': '',
'translations': hypotheses
}
reranked_hypotheses = reranker.rerank(hypotheses, reference)
assert reranked_hypotheses['translations'] == expected_output
def test_rerank_top1(hypotheses, reference, expected_best):
reranker = rerank.Reranker(metric="bleu", return_score=False)
reranked = reranker.rerank_top1(hypotheses, reference)
assert len(reranked.hypotheses
) == 1, "Rerank top1 should not return more than 1 hypothesis."
actual_hypothesis = reranked.hypotheses[0]
assert actual_hypothesis == expected_best
def test_rerank_top1_score(hypotheses, reference, expected_best,
expected_score):
reranker = rerank.Reranker(metric="bleu", return_score=True)
reranked_with_scores = reranker.rerank_top1(hypotheses, reference)
assert len(reranked_with_scores.hypotheses
) == 1, "Rerank top1 should not return more than 1 hypothesis."
actual_hypothesis = reranked_with_scores.hypotheses[0]
actual_score = reranked_with_scores.scores[0]
assert actual_hypothesis == expected_best
assert np.isclose(actual_score, expected_score)
def test_rerank_hypotheses_isometric(source, hypotheses, scores, reference,
expected_output, metric):
reranker = rerank.Reranker(metric=metric,
isometric_alpha=0.5,
return_score=False)
hypotheses = {
'sentence_id': 0,
'text': source,
'translation': '',
'scores': scores,
'translations': hypotheses
}
reranked_hypotheses = reranker.rerank(hypotheses, reference)
assert reranked_hypotheses['translations'] == expected_output
def test_rerank_return_score(source, hypotheses, reference, expected_scores):
reranker = rerank.Reranker(metric="bleu",
isometric_alpha=0.5,
return_score=True)
hypotheses = {
'sentence_id': 0,
'text': source,
'translation': '',
'translations': hypotheses
}
reranked_hypotheses = reranker.rerank(hypotheses, reference)
assert 'scores' in reranked_hypotheses
actual_scores = reranked_hypotheses['scores']
assert np.allclose(actual_scores, expected_scores)
