def test_serialize():
"""Test that models can correctly be saved to and loaded from gzipped json
"""
texts = data['texts']
classifier1 = load_model(os.path.join(TEST_MODEL_PATH, 'IR',
'IR_model.gz'))
temp_filename = os.path.join(SCRATCH_PATH, uuid.uuid4().hex)
classifier1.dump_model(temp_filename)
classifier2 = load_model(temp_filename)
classifier2.dump_model(temp_filename)
classifier3 = load_model(temp_filename)
preds1, preds2, preds3 = (classifier1.predict_proba(texts),
classifier2.predict_proba(texts),
classifier3.predict_proba(texts))
# Check that generated predictions are the same
assert np.array_equal(preds1, preds2)
assert np.array_equal(preds2, preds3)
# Check that model stats are the same
assert classifier1.stats == classifier2.stats == classifier3.stats
# Check that the calculated feature importance scores are the same
assert classifier1.feature_importances() == \
classifier2.feature_importances() == \
classifier3.feature_importances()
os.remove(temp_filename)
def test_disambiguate():
test_model = load_model(os.path.join(TEST_MODEL_PATH, 'IR',
'IR_model.gz'))
with open(os.path.join(TEST_MODEL_PATH, 'IR',
'IR_grounding_dict.json')) as f:
grounding_dict = json.load(f)
with open(os.path.join(TEST_MODEL_PATH, 'IR',
'IR_names.json')) as f:
names = json.load(f)
ad = AdeftDisambiguator(test_model, grounding_dict, names)
# case where there is a unique defining pattern
disamb1 = ad.disambiguate(example1)
assert disamb1[0] == 'HGNC:6091'
assert disamb1[1] == 'INSR'
assert disamb1[2]['HGNC:6091'] == 1.0
assert disamb1[2]['MESH:D011839'] == 0.0
# case where there are conflicting defining patterns
disamb2 = ad.disambiguate(example2)
preds = disamb2[2]
nonzero = {key for key, value in preds.items() if value > 0.0}
assert nonzero == {'HGNC:6091', 'MESH:D007333'}
# case without a defining pattern
disamb3 = ad.disambiguate(example3)
assert disamb3[0] == 'HGNC:6091'
assert disamb3[1] == 'INSR'
def test_serialize():
"""Test that models can correctly be saved to and loaded from gzipped json
"""
texts = data['texts']
classifier1 = load_model(os.path.join(TEST_MODEL_PATH, 'IR',
'IR_model.gz'))
temp_filename = os.path.join(SCRATCH_PATH, uuid.uuid4().hex)
classifier1.dump_model(temp_filename)
classifier2 = load_model(temp_filename)
classifier2.other_metadata = {'test': 'This is a test.'}
classifier2.dump_model(temp_filename)
classifier3 = load_model(temp_filename)
preds1, preds2, preds3 = (classifier1.predict_proba(texts),
classifier2.predict_proba(texts),
classifier3.predict_proba(texts))
# Check that generated predictions are the same
assert np.array_equal(preds1, preds2)
assert np.array_equal(preds2, preds3)
# Check that model stats are the same
assert classifier1.stats == classifier2.stats == classifier3.stats
# Check that the calculated feature importance scores are the same
assert classifier1.feature_importances() == \
classifier2.feature_importances() == \
classifier3.feature_importances()
# Check timestamps are unchanged
assert classifier1.timestamp == classifier2.timestamp == \
classifier3.timestamp
# Check hash of training set is unchanged
assert classifier1.training_set_digest == \
classifier2.training_set_digest == \
classifier3.training_set_digest
# Check standard deviations of feature values are unchanged
assert np.array_equal(classifier1._std, classifier2._std)
assert np.array_equal(classifier2._std, classifier3._std)
# Check classifier versions are unchanged
assert classifier1.version == classifier2.version == \
classifier3.version
# Check that model params are unchanged
assert classifier1.params == classifier2.params == classifier3.params
assert classifier2.other_metadata == classifier3.other_metadata
os.remove(temp_filename)
def load_disambiguator_directly(path):
"""Returns disambiguator located at path
Parameters
----------
path : str
Path to a disambiguation model. Must be a path to a directory
<model_name> containing the files
<model_name>_model.gz, <model_name>_grounding_dict.json, <model_name>_names.json
Returns
-------
py:class:`adeft.disambiguate.AdeftDisambiguator`
A disambiguation model loaded from folder specified by path
"""
model_name = os.path.basename(os.path.abspath(path))
model = load_model(os.path.join(path, model_name + '_model.gz'))
with open(os.path.join(path, model_name + '_grounding_dict.json')) as f:
grounding_dict = json.load(f)
with open(os.path.join(path, model_name + '_names.json')) as f:
names = json.load(f)
output = AdeftDisambiguator(model, grounding_dict, names)
return output
if os.path.isdir(model_path) and \
model_name in set(available_shortforms.values()):
names_path = os.path.join(model_path, f'{model_name}_names.json')
with open(names_path, 'r') as f:
names = strip_dictionary(json.load(f))
with open(names_path, 'w') as f:
json.dump(names, f)
gdict_path = os.path.join(model_path,
f'{model_name}_grounding_dict.json')
with open(gdict_path, 'r') as f:
grounding_dict = json.load(f)
grounding_dict = {shortform: strip_dictionary(grounding_map)
for shortform, grounding_map in
grounding_dict.items()}
model_file = os.path.join(model_path, f'{model_name}_model.gz')
model = load_model(model_file)
model.pos_labels = [label.strip() for label in model.pos_labels]
for i, label in (
enumerate(model.estimator.named_steps['logit'].classes_)):
model.estimator.named_steps['logit'].classes_[i] = \
label.strip()
model.dump_model(model_file)
with open(gdict_path, 'w') as f:
json.dump(grounding_dict, f)
for shortform in grounding_dict:
grounding_path = os.path.join(groundings_path, shortform)
names_path = os.path.join(grounding_path,
f'{shortform}_names.json')