def testMeanEncoder(self):
inputs = tf.concat([tf.ones([1, 5, 1]), 2 * tf.ones([1, 5, 1])], 0)
length = tf.constant([2, 4], dtype=tf.int32)
mask = tf.sequence_mask(length, maxlen=tf.shape(inputs)[1], dtype=inputs.dtype)
inputs *= tf.expand_dims(mask, -1)
encoder = encoders.MeanEncoder()
_, state, _ = encoder(inputs, sequence_length=length)
state = self.evaluate(state)
self.assertEqual(state[0][0], 1)
self.assertEqual(state[1][0], 2)
def testSequenceClassifier(self, mode):
model = models.SequenceClassifier(inputters.WordEmbedder(10),
encoders.MeanEncoder())
features_file, labels_file, data_config = self._makeToyClassifierData()
params = {"optimizer": "SGD", "learning_rate": 0.1}
self._testGenericModel(model,
mode,
features_file,
labels_file,
data_config,
prediction_heads=["classes"],
metrics=["accuracy"],
params=params)
def testSequenceTagger(self, mode):
model = models.SequenceTagger(inputters.WordEmbedder(10),
encoders.MeanEncoder(),
crf_decoding=True)
features_file, labels_file, data_config = self._makeToyTaggerData()
data_config["tagging_scheme"] = "bioes"
params = {"optimizer": "SGD", "learning_rate": 0.1}
self._testGenericModel(
model,
mode,
features_file,
labels_file,
data_config,
prediction_heads=["tags", "length"],
metrics=["accuracy", "precision", "recall", "f1"],
params=params)
def testSequenceClassifier(self, mode):
model = models.SequenceClassifier(
inputters.WordEmbedder("source_vocabulary", 10),
encoders.MeanEncoder(), "target_vocabulary")
features_file, labels_file, metadata = self._makeToyClassifierData()
params = {
"optimizer": "GradientDescentOptimizer",
"learning_rate": 0.1
}
self._testGenericModel(model,
mode,
features_file,
labels_file,
metadata,
prediction_heads=["classes"],
metrics=["accuracy"],
params=params)
def testSequenceTagger(self, mode):
model = models.SequenceTagger(
inputters.WordEmbedder("source_vocabulary", 10),
encoders.MeanEncoder(),
"target_vocabulary",
crf_decoding=True,
tagging_scheme="bioes")
features_file, labels_file, metadata = self._makeToyTaggerData()
params = {
"optimizer": "GradientDescentOptimizer",
"learning_rate": 0.1}
self._testGenericModel(
model,
mode,
features_file,
labels_file,
metadata,
prediction_heads=["tags", "length"],
metrics=["accuracy", "precision", "recall", "f1"],
params=params)
