def test_store_classifier_data(self):
"""Test the store_classifier_data method."""
exp_id = u'1'
next_scheduled_check_time = datetime.datetime.utcnow()
state_name = 'Home'
interaction_id = 'TextInput'
job_id = self._create_classifier_training_job(
feconf.INTERACTION_CLASSIFIER_MAPPING['TextInput']['algorithm_id'],
interaction_id, exp_id, 1, next_scheduled_check_time, [],
state_name, feconf.TRAINING_JOB_STATUS_PENDING, {}, 1)
# Retrieve classifier data from GCS and ensure that content is same.
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
classifier_data = (
self._get_classifier_data_from_classifier_training_job(
classifier_training_job))
self.assertEqual(json.loads(classifier_data.model_json), {})
classifier_data_proto = text_classifier_pb2.TextClassifierFrozenModel()
classifier_data_proto.model_json = json.dumps(
{'classifier_data': 'data'})
classifier_services.store_classifier_data(job_id,
classifier_data_proto)
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
classifier_data = (
self._get_classifier_data_from_classifier_training_job(
classifier_training_job))
self.assertDictEqual(json.loads(classifier_data.model_json),
{'classifier_data': 'data'})
def post(self):
"""Handles POST requests."""
payload_proto = (
training_job_response_payload_pb2.TrainingJobResponsePayload())
payload_proto.ParseFromString(self.request.body)
if not validate_job_result_message_proto(payload_proto.job_result):
raise self.InvalidInputException
job_id = payload_proto.job_result.job_id
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
if classifier_training_job.status == (
feconf.TRAINING_JOB_STATUS_FAILED):
# Send email to admin and admin-specified email recipients.
# Other email recipients are specified on admin config page.
email_manager.send_job_failure_email(job_id)
raise self.InternalErrorException(
'The current status of the job cannot transition to COMPLETE.')
classifier_data_proto = getattr(
payload_proto.job_result,
payload_proto.job_result.WhichOneof('classifier_frozen_model'))
classifier_services.store_classifier_data(
job_id, classifier_data_proto)
# Update status of the training job to 'COMPLETE'.
classifier_services.mark_training_job_complete(job_id)
return self.render_json({})
def post(self):
"""Handles POST requests."""
signature = self.payload.get('signature')
message = self.payload.get('message')
vm_id = self.payload.get('vm_id')
if vm_id == feconf.DEFAULT_VM_ID and not feconf.DEV_MODE:
raise self.UnauthorizedUserException
if not validate_job_result_message_dict(message):
raise self.InvalidInputException
if not verify_signature(message, vm_id, signature):
raise self.UnauthorizedUserException
job_id = message['job_id']
classifier_data = message['classifier_data']
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
if classifier_training_job.status == (
feconf.TRAINING_JOB_STATUS_FAILED):
raise self.InternalErrorException(
'The current status of the job cannot transition to COMPLETE.')
try:
classifier_services.store_classifier_data(job_id, classifier_data)
except Exception as e:
raise self.InternalErrorException(e)
# Update status of the training job to 'COMPLETE'.
classifier_services.mark_training_job_complete(job_id)
return self.render_json({})
def post(self):
"""Handles POST requests."""
signature = self.payload.get('signature')
message = self.payload.get('message')
vm_id = self.payload.get('vm_id')
if vm_id == feconf.DEFAULT_VM_ID and not constants.DEV_MODE:
raise self.UnauthorizedUserException
if not validate_job_result_message_dict(message):
raise self.InvalidInputException
if not verify_signature(message, vm_id, signature):
raise self.UnauthorizedUserException
job_id = message['job_id']
# The classifier data received in the payload has all floating point
# values stored as strings. This is because floating point numbers
# are represented differently on GAE(Oppia) and GCE(Oppia-ml).
# Therefore, converting all floating point numbers to string keeps
# signature consistent on both Oppia and Oppia-ml.
# For more info visit: https://stackoverflow.com/q/40173295
classifier_data = (
classifier_services.
convert_strings_to_float_numbers_in_classifier_data( #pylint: disable=line-too-long
message['classifier_data_with_floats_stringified']))
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
if classifier_training_job.status == (
feconf.TRAINING_JOB_STATUS_FAILED):
# Send email to admin and admin-specified email recipients.
# Other email recipients are specified on admin config page.
email_manager.send_job_failure_email(job_id)
raise self.InternalErrorException(
'The current status of the job cannot transition to COMPLETE.')
try:
classifier_services.store_classifier_data(job_id, classifier_data)
except Exception as e:
raise self.InternalErrorException(e)
# Update status of the training job to 'COMPLETE'.
classifier_services.mark_training_job_complete(job_id)
return self.render_json({})
def test_creation_of_state_classifier_mapping(self):
super(ExplorationStateClassifierMappingTests, self).setUp()
exploration_id = '15'
self.login(self.VIEWER_EMAIL)
self.signup(self.VIEWER_EMAIL, self.VIEWER_USERNAME)
exp_services.delete_demo(exploration_id)
# We enable ENABLE_ML_CLASSIFIERS so that the subsequent call to
# save_exploration handles job creation for trainable states.
# Since only one demo exploration has a trainable state, we modify our
# values for MIN_ASSIGNED_LABELS and MIN_TOTAL_TRAINING_EXAMPLES to let
# the classifier_demo_exploration.yaml be trainable. This is
# completely for testing purposes.
with self.swap(feconf, 'ENABLE_ML_CLASSIFIERS', True):
with self.swap(feconf, 'MIN_TOTAL_TRAINING_EXAMPLES', 5):
with self.swap(feconf, 'MIN_ASSIGNED_LABELS', 1):
exp_services.load_demo(exploration_id)
# Retrieve job_id of created job (because of save_exp).
all_jobs = classifier_models.ClassifierTrainingJobModel.get_all()
self.assertEqual(all_jobs.count(), 1)
for job in all_jobs:
job_id = job.id
classifier_services.store_classifier_data(job_id, {})
expected_state_classifier_mapping = {
'text': {
'algorithm_id': 'LDAStringClassifier',
'classifier_data': {},
'data_schema_version': 1
}
}
# Call the handler.
exploration_dict = self.get_json(
'%s/%s' % (feconf.EXPLORATION_INIT_URL_PREFIX, exploration_id))
retrieved_state_classifier_mapping = exploration_dict[
'state_classifier_mapping']
self.assertEqual(expected_state_classifier_mapping,
retrieved_state_classifier_mapping)
def test_store_classifier_data(self):
"""Test the store_classifier_data method."""
exp_id = u'1'
next_scheduled_check_time = datetime.datetime.utcnow()
state_name = 'Home'
interaction_id = 'TextInput'
job_id = classifier_models.ClassifierTrainingJobModel.create(
feconf.INTERACTION_CLASSIFIER_MAPPING['TextInput']['algorithm_id'],
interaction_id, exp_id, 1, next_scheduled_check_time, [],
state_name, feconf.TRAINING_JOB_STATUS_PENDING, None, 1)
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
self.assertEqual(classifier_training_job.classifier_data, None)
classifier_services.store_classifier_data(job_id, {})
classifier_training_job = (
classifier_services.get_classifier_training_job_by_id(job_id))
self.assertEqual(classifier_training_job.classifier_data, {})
def test_can_not_store_classifier_data_due_to_invalid_job_id(self):
with self.assertRaisesRegexp(
Exception,
'The ClassifierTrainingJobModel corresponding to the '
'job_id of the ClassifierTrainingJob does not exist.'):
classifier_services.store_classifier_data('invalid_job_id', {})
