def loadModels(self, force=False):
config = RasaNLUConfig(self.default_config_file)
slotConfig = RasaNLUConfig(self.default_config_file_slots)
#print("start load modesls")
# use the files as templates and override config on top for each model
for model in self.models:
config.override({"project": "nlu", "fixed_model_name": model})
slotConfig.override({
"project": "nlu",
"fixed_model_name": "{}_slots".format(model)
})
if not self.isNluModelMissing(model):
#print("model not missing")
#if (self.hasTrainingMaterials(model)):
#self.sendTrainingRequest(model)
#else:
#print('missing model and training data')
#else:
## load trained model
# allow that training id is None for load existing model
if (self.isNluModelModified(model) or force):
print("loading nlu model {}".format(model))
self.nlu_model_modified[model] = self.getNluModelModified(
model)
self.nlu_modified[model] = self.getNluModified(model)
self.interpreter[model] = Interpreter.load(
"{}/{}/".format(self.nlu_model_path, model), config)
self.interpreter_slots[model] = Interpreter.load(
"{}/{}_slots/".format(self.nlu_model_path, model),
slotConfig)
print('loaded nlu model {}'.format(model))
if model in self.trainingIds and self.trainingIds[
model] is not None:
self.training_client.unsubscribe(
"hermes/training/complete/{}".format(
self.trainingIds[model]))
self.training_request_models[
self.trainingIds[model]] = None
self.trainingIds[model] = None
def do_training(self, run_event):
while True and run_event.is_set():
#print('do training')
if len(self.queue) > 0:
msg = self.queue.pop(0)
payload = json.loads(msg.payload)
theId = payload.get('id')
print('HANDLE TRAINING REQUEST {}'.format(theId))
if theId is not None:
# .decode('utf-8')
trainingType = payload.get('type')
# DO THE TRAINING
if trainingType == "rasanlu":
# set variables from msg payload or defaults
trainingConfig = payload.get('config')
#if trainingConfig is None:
#with open('rasa_config/config.json', 'r') as content_file:
#trainingConfig = content_file.read()
trainingConfig = json.loads(trainingConfig)
slotConfig = payload.get('config_slots')
#if slotConfig is None:
#with open('rasa_config/config-slots.json', 'r') as content_file:
#slotConfig = content_file.read()
slotConfig = json.loads(slotConfig)
project = payload.get('project', 'nlu')
model = payload.get('model', 'default')
training_data = payload.get('examples', '')
tmpdir = tempfile.mkdtemp()
# train intents and slots
rasaConfig = RasaNLUConfig()
rasaConfig.override(trainingConfig)
trainer = Trainer(rasaConfig)
data = SnipsMarkdownToJson(training_data)
trainer.train(
TrainingData(
data.common_examples,
get_entity_synonyms_dict(
data.entity_synonyms)))
model_directory = trainer.persist(
tmpdir,
project_name=project,
fixed_model_name=model)
# train slots only for partial query
rasaConfig.override(slotConfig)
trainer = Trainer(rasaConfig)
data = SnipsMarkdownToJson(training_data)
trainer.train(
TrainingData(
data.common_examples,
get_entity_synonyms_dict(
data.entity_synonyms)))
model_directory = trainer.persist(
tmpdir,
project_name=project,
fixed_model_name="{}_slots".format(model))
# send mqtt training/complete
self.send_training_complete(
theId, "{}/{}".format(tmpdir, project))
#shutil.rmtree(tmpdir)
elif trainingType == "rasacore":
print("training rasa core")
training_data = payload.get(
'stories') #,SAMPLE_STORIES)
trainingFile = tempfile.NamedTemporaryFile(
delete=False, suffix='.md')
trainingFile.write(training_data)
trainingFile.close()
#print((training_data))
domain_data = payload.get('domain')
#if domain_data is None:
#with open('rasa_config/domain.yml', 'r') as content_file:
#domain_data = content_file.read()
#print(domain_data)
tmpdir = tempfile.mkdtemp()
domainFile = io.open(
os.path.join(tmpdir, "domain.yml"), "w")
#domainFile = tempfile.NamedTemporaryFile(delete = False,suffix='.yml')
domainFile.write(domain_data)
domainFile.close()
agent = SnipsMqttAgent.createAgent(tmpdir)
#agent = Agent(domainFile.name,policies=[MemoizationPolicy(), KerasPolicy()])
agent.train(
trainingFile.name,
max_history=3,
epochs=30, #was 100
batch_size=50,
augmentation_factor=50,
validation_split=0.2)
print("traing rasa core done train")
# cleanup
domainFile.close()
os.unlink(domainFile.name)
os.unlink(trainingFile.name)
# persist
modelPath = tempfile.mkdtemp()
agent.persist(modelPath)
self.send_training_complete(theId, modelPath)
print("traing rasa core sent")
elif trainingType == "kaldi":
pass
elif trainingType == "piwho":
pass
elif trainingType == "snowboy":
pass
# I WISH
#elif trainingType == "snips":
#pass
else:
print("Required ID missing in training request")
time.sleep(5)
