def test_retrieval_of_classifier_training_jobs_from_exploration_attributes(
self):
"""Test the get_classifier_training_job method."""
exp_id = u'1'
next_scheduled_check_time = datetime.datetime.utcnow()
state_name = u'टेक्स्ट'
algorithm_id = feconf.INTERACTION_CLASSIFIER_MAPPING['TextInput'][
'algorithm_id']
algorithm_version = feconf.INTERACTION_CLASSIFIER_MAPPING['TextInput'][
'algorithm_version']
job_id = self._create_classifier_training_job(
algorithm_id, 'TextInput', exp_id, 1, next_scheduled_check_time,
[], state_name, feconf.TRAINING_JOB_STATUS_NEW, {},
algorithm_version)
classifier_models.StateTrainingJobsMappingModel.create(
exp_id, 1, state_name, {algorithm_id: job_id})
classifier_training_job = (
classifier_services.get_classifier_training_job(
exp_id, 1, state_name, algorithm_id))
self.assertIsNotNone(classifier_training_job)
self.assertEqual(classifier_training_job.exp_id, exp_id)
self.assertEqual(classifier_training_job.exp_version, 1)
self.assertEqual(classifier_training_job.state_name, state_name)
self.assertEqual(classifier_training_job.job_id, job_id)
# Test that method returns a list with None as elements when job does
# not exist.
false_state_name = 'false_name'
classifier_training_job = (
classifier_services.get_classifier_training_job(
exp_id, 1, false_state_name, algorithm_id))
self.assertIsNone(classifier_training_job)
def test_reverted_exploration_maintains_classifier_model_mapping(
self) -> None:
"""Test if the classifier model mapping is maintained when an
exploration is reverted.
"""
state_name = 'Home'
exploration = exp_fetchers.get_exploration_by_id(self.exp_id)
interaction_id = exploration.states[state_name].interaction.id
algorithm_id = feconf.INTERACTION_CLASSIFIER_MAPPING[interaction_id][
'algorithm_id']
# Make changes to the exploration.
change_list = [
exp_domain.ExplorationChange({
'cmd': exp_domain.CMD_EDIT_EXPLORATION_PROPERTY,
'property_name': 'title',
'new_value': 'A new title'
})
]
with self.swap(feconf, 'ENABLE_ML_CLASSIFIERS', True):
exp_services.update_exploration( # type: ignore[no-untyped-call]
feconf.SYSTEM_COMMITTER_ID, self.exp_id, change_list, '')
current_exploration = exp_fetchers.get_exploration_by_id(self.exp_id)
# Store the job before reverting the exploration.
old_job = classifier_services.get_classifier_training_job(
self.exp_id, current_exploration.version, state_name, algorithm_id)
# Ruling out the possibility of None for mypy type checking.
assert old_job is not None
old_job_id = old_job.job_id
# Revert the exploration.
with self.swap(feconf, 'ENABLE_ML_CLASSIFIERS', True):
exp_services.revert_exploration( # type: ignore[no-untyped-call]
feconf.SYSTEM_COMMITTER_ID, self.exp_id,
current_exploration.version, current_exploration.version - 1)
# Verify if classifier model mapping is maintained using the job ID.
reverted_exploration = exp_fetchers.get_exploration_by_id(self.exp_id)
self.assertEqual(reverted_exploration.version,
current_exploration.version + 1)
new_job = classifier_services.get_classifier_training_job(
self.exp_id, reverted_exploration.version, state_name,
algorithm_id)
# Ruling out the possibility of None for mypy type checking.
assert new_job is not None
new_job_id = new_job.job_id
self.assertEqual(old_job_id, new_job_id)
def test_trained_classifier_handler(self):
# Normal end-to-end test.
self.post_blob('/ml/trainedclassifierhandler',
self.payload_proto.SerializeToString(),
expected_status_int=200)
classifier_training_job = (
classifier_services.get_classifier_training_job(
self.exp_id, self.exploration.version, 'Home',
self.algorithm_id))
self.assertIsNotNone(classifier_training_job)
classifier_data = (
self._get_classifier_data_from_classifier_training_job(
classifier_training_job))
self.assertEqual(json.loads(classifier_data.model_json),
self.classifier_data)
self.assertEqual(classifier_training_job.status,
feconf.TRAINING_JOB_STATUS_COMPLETE)
def test_get_trained_classifier_handler(self):
self.post_blob('/ml/trainedclassifierhandler',
self.payload_proto.SerializeToString(),
expected_status_int=200)
classifier_training_job = (
classifier_services.get_classifier_training_job(
self.exp_id, self.exploration.version, 'Home',
self.algorithm_id))
params = {
'exploration_id': self.exp_id,
'exploration_version': self.exploration.version,
'state_name': 'Home',
}
response = self.get_json('/ml/trainedclassifierhandler',
params=params,
expected_status_int=200)
self.assertEqual(response['gcs_filename'],
classifier_training_job.classifier_data_filename)
def setUp(self):
super(TrainedClassifierHandlerTests, self).setUp()
self.exp_id = 'exp_id1'
self.title = 'Testing Classifier storing'
self.category = 'Test'
yaml_path = os.path.join(feconf.TESTS_DATA_DIR,
'string_classifier_test.yaml')
with python_utils.open_file(yaml_path, 'r') as yaml_file:
self.yaml_content = yaml_file.read()
self.signup(self.CURRICULUM_ADMIN_EMAIL,
self.CURRICULUM_ADMIN_USERNAME)
self.signup('*****@*****.**', 'mod')
assets_list = []
with self.swap(feconf, 'ENABLE_ML_CLASSIFIERS', True):
exp_services.save_new_exploration_from_yaml_and_assets(
feconf.SYSTEM_COMMITTER_ID, self.yaml_content, self.exp_id,
assets_list)
self.exploration = exp_fetchers.get_exploration_by_id(self.exp_id)
self.algorithm_id = feconf.INTERACTION_CLASSIFIER_MAPPING[
self.exploration.states['Home'].interaction.id]['algorithm_id']
self.algorithm_version = feconf.INTERACTION_CLASSIFIER_MAPPING[
self.exploration.states['Home'].interaction.
id]['algorithm_version']
self.classifier_data = {
'_alpha': 0.1,
'_beta': 0.001,
'_prediction_threshold': 0.5,
'_training_iterations': 25,
'_prediction_iterations': 5,
'_num_labels': 10,
'_num_docs': 12,
'_num_words': 20,
'_label_to_id': {
'text': 1
},
'_word_to_id': {
'hello': 2
},
'_w_dp': [],
'_b_dl': [],
'_l_dp': [],
'_c_dl': [],
'_c_lw': [],
'_c_l': [],
}
classifier_training_job = (
classifier_services.get_classifier_training_job(
self.exp_id, self.exploration.version, 'Home',
self.algorithm_id))
self.assertIsNotNone(classifier_training_job)
self.job_id = classifier_training_job.job_id
# TODO(pranavsid98): Replace the three commands below with
# mark_training_job_pending after Giritheja's PR gets merged.
classifier_training_job_model = (
classifier_models.ClassifierTrainingJobModel.get(self.job_id,
strict=False))
classifier_training_job_model.status = (
feconf.TRAINING_JOB_STATUS_PENDING)
classifier_training_job_model.update_timestamps()
classifier_training_job_model.put()
self.job_result = (training_job_response_payload_pb2.
TrainingJobResponsePayload.JobResult())
self.job_result.job_id = self.job_id
classifier_frozen_model = (
text_classifier_pb2.TextClassifierFrozenModel())
classifier_frozen_model.model_json = json.dumps(self.classifier_data)
self.job_result.text_classifier.CopyFrom(classifier_frozen_model)
self.payload_proto = (
training_job_response_payload_pb2.TrainingJobResponsePayload())
self.payload_proto.job_result.CopyFrom(self.job_result)
self.payload_proto.vm_id = feconf.DEFAULT_VM_ID
self.secret = feconf.DEFAULT_VM_SHARED_SECRET
self.payload_proto.signature = classifier_services.generate_signature(
python_utils.convert_to_bytes(self.secret),
python_utils.convert_to_bytes(
self.payload_proto.job_result.SerializeToString()),
self.payload_proto.vm_id)
self.payload_for_fetching_next_job_request = {
'vm_id': feconf.DEFAULT_VM_ID,
'message': json.dumps({})
}
self.payload_for_fetching_next_job_request['signature'] = (
classifier_services.generate_signature(
python_utils.convert_to_bytes(self.secret),
python_utils.convert_to_bytes(
self.payload_for_fetching_next_job_request['message']),
self.payload_for_fetching_next_job_request['vm_id']))
