def json(self):
if self.training_event_log is not None:
event_log = self.training_event_log.name
else:
event_log = self.file_name.split('_')[0]
return {
'id': self.id,
'creationDate': self.creation_date,
'algorithm': self.algorithm,
'trainingEventLog': event_log,
'trainingDuration': self.training_duration,
'fileName': self.file_name,
'trainingHost': self.training_host,
'hyperparameters':
json.loads(self.hyperparameters.replace("'", '"')),
'modelFileExists': ModelFile(self.file_name).path.exists(),
'cached': ModelFile(self.file_name).result_file.exists()
}
def load(self, file_name):
# load model file
file_name = ModelFile(file_name)
# load model
from keras.models import load_model
from keras.utils import CustomObjectScope
from april.anomalydetection.binet.attention import Attention
with CustomObjectScope({'Attention': Attention}):
self._model = load_model(file_name.str_path)
def save(self, file_name=None):
"""Save the class instance using pickle.
:param file_name: Custom file name
:return: the file path
"""
if self._model is not None:
model_file = ModelFile(file_name)
self._save(model_file.str_path)
return model_file
else:
raise RuntimeError(
'Saving not possible. No model has been trained yet.')
def load(self, file_name):
"""
Load a class instance from a pickle file. If no extension or absolute path are given the method assumes the
file to be located inside the models dir. Model extension can be omitted in the file name.
:param file_name: Path to saved model file.
:return: None
"""
# load model file
model_file = ModelFile(file_name)
# load model
self._model = pickle.load(open(model_file.path, 'rb'))
def __init__(self, model):
if not isinstance(model, ModelFile):
self.model = ModelFile(model)
else:
self.model = model
self.model_file = self.model.path
self.model_name = self.model.name
self.eventlog_name = self.model.event_log_name
self.process_model_name = self.model.model
self.noise = self.model.p
self.dataset_id = self.model.id
self.model_date = self.model.date
self.ad_ = AD.get(self.model.ad)()
self._dataset = None
self._result = None
self._binarizer = None
self._event_log_df = None
self._classification = None
import warnings
warnings.filterwarnings("ignore", category=UndefinedMetricWarning)