def to_method_object(self):
"""Convert the enum to an instance of `BaselineMethod`."""
if self == self.TF_IDF:
return keyword_based.TfIdfMethod()
elif self == self.BM25:
return keyword_based.BM25Method()
elif self == self.USE_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder/2"))
elif self == self.USE_LARGE_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder-large/3"))
elif self == self.ELMO_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/elmo/1"))
elif self == self.USE_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder/2"))
elif self == self.USE_LARGE_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder-large/3"))
elif self == self.ELMO_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/elmo/1"))
raise ValueError("Unknown method {}".format(self))
def test_train_then_rank(self):
mock_encoder = mock.Mock()
def _random_encode(texts):
return np.random.normal(size=(len(texts), 3))
mock_encoder.encode.side_effect = _random_encode
method = vector_based.VectorMappingMethod(mock_encoder,
learning_rates=[1],
regularizers=[0])
# Use 104 elements, so that the encoding must be batched.
method.train(["context"] * 104, ["response"] * 104)
mock_encoder.encode.assert_has_calls([
mock.call(["context"] * 100),
mock.call(["response"] * 100),
mock.call(["context"] * 4),
mock.call(["response"] * 4),
])
assignments = method.rank_responses(["x", "y", "z"], ["a", "b", "c"])
self.assertEqual((3, ), assignments.shape)
for id_ in assignments:
self.assertGreaterEqual(id_, 0)
self.assertLess(id_, 3)
def to_method_object(self):
"""Convert the enum to an instance of `BaselineMethod`."""
if self == self.TF_IDF:
return keyword_based.TfIdfMethod()
elif self == self.BM25:
return keyword_based.BM25Method()
elif self == self.USE_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder/2"))
elif self == self.USE_LARGE_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder-large/3"))
elif self == self.ELMO_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/elmo/1"))
elif self == self.USE_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder/2"))
elif self == self.USE_LARGE_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder-large/3"))
elif self == self.ELMO_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/elmo/1"))
elif self == self.BERT_SMALL_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.BERTEncoder(
"https://tfhub.dev/google/"
"bert_uncased_L-12_H-768_A-12/1"))
elif self == self.BERT_SMALL_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.BERTEncoder(
"https://tfhub.dev/google/"
"bert_uncased_L-12_H-768_A-12/1"))
elif self == self.BERT_LARGE_SIM:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.BERTEncoder(
"https://tfhub.dev/google/"
"bert_uncased_L-24_H-1024_A-16/1"))
elif self == self.BERT_LARGE_MAP:
return vector_based.VectorMappingMethod(
encoder=vector_based.BERTEncoder(
"https://tfhub.dev/google/"
"bert_uncased_L-24_H-1024_A-16/1"))
elif self == self.USE_QA:
return vector_based.VectorSimilarityMethod(
encoder=vector_based.TfHubEncoder(
"https://tfhub.dev/google/"
"universal-sentence-encoder-multilingual-qa/1",
is_dual=True))
raise ValueError("Unknown method {}".format(self))