def test_preprocess_transform_returns_vectors_of_given_max_len(self):
model = ElmoModel(
max_len=50,
fasttext_model=self.fasttext_model,
elmo_embedder=self.elmo_embedder,
rnn_units = 100, bidirectional=False,
**self.model_args,
)
X = ["Esto no es agresivo", "Esto sí es agresivo"]
model.preprocess_fit(X)
ret = model.preprocess_transform(X)
self.assertEqual(len(ret), 2)
self.assertEqual(ret[0].shape, (2, 50, self.elmo_dim))
self.assertEqual(ret[1].shape, (2, 50, self.ft_dim))
def test_it_is_created_with_both(self):
model = ElmoModel(
max_len=50,
fasttext_model=self.fasttext_model,
elmo_embedder=self.elmo_embedder,
**self.model_args,
)
self.assertEqual(model.layers[3].input_shape,
(None, 50, self.ft_dim + self.elmo_dim))
def test_it_creates_with_embedding_size(self):
model = ElmoModel(
max_len=50,
fasttext_model=self.fasttext_model,
elmo_embedder=self.elmo_embedder,
rnn_units = 100, bidirectional=False,
**self.model_args,
)
self.assertEqual(
model.layers[4].output_shape,
(None, 100)
)
def test_with_real_embedder(self):
model = ElmoModel(
max_len=50,
fasttext_model=None,
elmo_embedder=Embedder("models/elmo/es/"),
**self.model_args,
)
X = ["Esto no es agresivo", "Esto sí es agresivo"]
y = np.array([0, 1]).reshape(-1, 1)
model.compile(loss='binary_crossentropy', optimizer='adam',
metrics=['accuracy'])
model.fit(X, y, epochs=2)
def create_model(params, embedder):
params = params.copy()
K.clear_session()
dropout = params.pop('dropout')
recursive_class = params.pop('recursive_class')
dense_last_layer = params.pop('dense_last_layer')
char_model = CharModel(vocab_size=params.pop('char__vocab_size'),
max_charlen=params.pop('char__max_charlen'),
embedding_dim=params.pop('char__embedding_dim'),
tokenize_args={
"stem": params.pop('char__stem'),
"alpha_only": params.pop('char__alpha_only'),
},
filters=params.pop('char__filters'),
kernel_size=params.pop('char__kernel_size'),
pooling_size=params.pop('char__pooling_size'),
dense_units=dense_last_layer,
recursive_class=recursive_class,
dropout=dropout)
elmo_model = ElmoModel(
max_len=50,
embedder=embedder,
lstm_units=params.pop('elmo__lstm_units'),
tokenize_args={'deaccent': params.pop('elmo__deaccent')},
dense_units=dense_last_layer,
recursive_class=recursive_class,
dropout=dropout)
bow_model = BowModel(
num_words=params.pop('bow__num_words'),
dense_units=[1024, dense_last_layer],
dropout=dropout,
)
merge_model = MergeModel([char_model, elmo_model, bow_model])
optimizer_args = {"lr": params.pop('lr'), "decay": params.pop('decay')}
merge_model.compile(loss='binary_crossentropy',
optimizer=Adam(**optimizer_args),
metrics=['accuracy'])
assert (len(params) == 0)
return merge_model
def test_it_can_be_fitted(self):
model = ElmoModel(
max_len=50,
fasttext_model=self.fasttext_model,
elmo_embedder=self.elmo_embedder,
rnn_units = 100, bidirectional=False,
**self.model_args,
)
X = ["Esto no es agresivo", "Esto sí es agresivo"]
y = np.array([0, 1]).reshape(-1, 1)
model.compile(loss='binary_crossentropy', optimizer='adam',
metrics=['accuracy'])
model.fit(X, y, epochs=2)