def test_fit_predict_dump_load(self):
"""Fits a model, tests that it predicts correctly, dumps and loads it, then tests that it still predicts"""
model = Model(self.pipeline)
# Test attempting to predict before fitting
with self.assertRaises(RuntimeError):
model.predict('Lorem ipsum dolor sit amet.')
model.fit(self.dataset,
groundtruth_directory=self.groundtruth_2_directory)
# Test X and y data are set
self.assertTrue(model.X_data)
self.assertTrue(model.y_data)
# Test that there is at least one prediction
resulting_ann = model.predict(
'To exclude the possibility that alterations in PSSD might be a consequence of changes in the volume of reference, we used a subset of the vibratome sections'
)
self.assertIsInstance(resulting_ann, Annotations)
self.assertTrue(resulting_ann)
# Test prediction over directory
resulting_dataset = model.predict(
self.dataset.data_directory,
prediction_directory=self.prediction_directory)
self.assertIsInstance(resulting_dataset, Dataset)
self.assertEqual(len(self.dataset), len(resulting_dataset))
# Test that groundtruth is written
groundtruth_dataset = Dataset(self.groundtruth_2_directory)
expected = [d.file_name for d in self.dataset]
actual = [d.file_name for d in groundtruth_dataset]
self.assertListEqual(expected, actual)
# Test that the groundtruth ann files have content
for ann in groundtruth_dataset.generate_annotations():
self.assertTrue(ann)
# Test pickling a model
pickle_path = os.path.join(self.prediction_directory, 'test.pkl')
model.dump(pickle_path)
new_model = Model(self.pipeline)
new_model.load(pickle_path)
# Test that there is at least one prediction
resulting_ann = new_model.predict(
'To exclude the possibility that alterations in PSSD might be a consequence of changes in the volume of reference, we used a subset of the vibratome sections'
)
self.assertIsInstance(resulting_ann, Annotations)
self.assertTrue(resulting_ann)
def test_prediction_with_testing_pipeline(self):
"""Tests that a model created with the BiLSTM+CRF can be fitted and used to predict"""
pipeline = LstmSystematicReviewPipeline(
entities=self.entities,
word_embeddings=word_embeddings,
cuda_device=cuda_device)
model = Model(pipeline)
model.fit(self.dataset)
resulting_dataset = model.predict(
self.dataset, prediction_directory=self.prediction_directory)
self.assertIsInstance(resulting_dataset, Dataset)
# Test that there is at least one prediction
if not any(resulting_dataset.generate_annotations()):
warn("The model did not generate any predictions")
def test_cross_validate_fit_predict(self):
"""Tests that a model created with BERT can be fitted and used to predict, with and without the CRF layer"""
pipeline = BertPipeline(entities=self.entities,
pretrained_model='bert-base-cased',
batch_size=self.batch_size,
cuda_device=cuda_device)
pipeline_crf = BertPipeline(entities=self.entities,
pretrained_model='bert-base-cased',
batch_size=self.batch_size,
cuda_device=cuda_device,
using_crf=True)
for pipe in [pipeline, pipeline_crf]:
model = Model(pipe)
model.cross_validate(self.dataset, 2)
model.fit(self.dataset)
resulting_dataset = model.predict(
self.dataset, prediction_directory=self.prediction_directory)
self.assertIsInstance(resulting_dataset, Dataset)
# Test that there is at least one prediction
if not any(resulting_dataset.generate_annotations()):
warn("The model did not generate any predictions")
# Selects the pre-processing pipeline this model should be trained with respect to.
pipeline = SystematicReviewPipeline(entities=entities, use_metamap=True)
model = Model(pipeline, n_jobs=1)
# number of cores to utilize during feature extraction when training the model.
# Note: this is done by forking, not threading hence utlizes a large amount of memory.
# Write information about model before training
with open(model_directory + "/model_information.txt", 'w') as model_info:
model_info.write("Entities: [%s]\n" % ", ".join(entities))
model_info.write("Training Files: %i\n" %
len(train_dataset.get_data_files()))
model_info.write(model_notes + "\n")
model_info.write(str(model))
model.fit(train_dataset)
# dump fitted model
current_time = datetime.datetime.fromtimestamp(
time.time()).strftime('%Y_%m_%d_%H.%M.%S')
model.dump(model_directory + "/tac_2018_%s_%s.pkl" %
(model_name, current_time))
# predicts over the datasets in evaluation_dataset utilizing the model trained above, then stores those predictions
# in a given output directory
model.predict(evaluation_dataset,
prediction_directory=os.path.join(model_directory,
'predictions'))
# performs sequence stratified cross validation over the trained model.
# Note that all extracted features are stored in memory while this runs.