Training and Loading Multiple Models From Same Project

Rasa Open Source
1

My aim is to train different models from the same project.

Basically I am having two different models with different intents i.e Weather and Restaurent models.

How can i train and load this models. Thanks in Adavance.

2

@Quresh_Abrar you can train different models by specifying project name, and you can load them in server. by specify the project name, you can use them.

python -m rasa_nlu.server -c nlu.yml --path models/nlu --port 5003

we can use any of these.

1-http://localhost:5003/parse?q=hi&project=resturant
2- http://localhost:5003/parse?q=hi&project=hotel

it is possible for nlu only. I am looking the same for rasa core. But I did not get the solution up till now

3

Thank you @azizullah2017 for your reply.

you explained about loading the models,

But How to train multiple models in a same project. and what will be the configuration files for multiple models, and should i need to specify separate yml file for each or how ?

4

are you asking about rasa nlu or rasa core ?

5

@azizullah2017 rasa nlu

6

$ python -m rasa_nlu.train
–config config.yml
–data nlu_data/
–path projects

https://rasa.com/docs/nlu/master/dataformat/#section-dataformat

7

@azizullah2017 Thank you

1 Like
8

Even I am looking for the same in rasa core. Please tell me if you find anything.

9

Rasa core does not support multiple models. You have to write your custom code. you have to create multiple agents this is my code

def train_dialogue(domain_file = 'domain.yml',
					model_path = './models/dialogue',
					training_data_file = './data/stories.md'):
					
	agent = Agent(domain_file, policies = [MemoizationPolicy(), KerasPolicy(max_history=3, epochs=200, batch_size=50)])
	data = agent.load_data(training_data_file)	
	

	agent.train(data)
				
	agent.persist(model_path)
	return agent
	
def run_weather_bot(serve_forever=True):
	interpreter = RasaNLUInterpreter('./models/nlu/Auto/Auto')
	domain = TemplateDomain.load("models/dialogue/domain.yml")
	action_endpoint = EndpointConfig(url="http://localhost:5055/webhook")
        tracker_store = RedisTrackerStore(domain,host="localhost", port=6379, db= 13)

	agent = Agent.load('domain', interpreter=interpreter,tracker_store=tracker_store, action_endpoint=action_endpoint)
	rasa_core.run.serve_application(agent ,channel='cmdline')
		
	return agent
	
if __name__ == '__main__':
	run_weather_bot()

you can get help

4 Likes
10

Hi, I’m very new in Rasa. I’m not getting that how can we give custom names to the models, trained and saved as zip|tar folder inside models directory? And, Is there any other way to train multiple models withing same project using mysql and without using Redis?

1 Like
11 
rasa train \
--config config/en/config.yml \
--domain config/en/domain.yml \
--data data/en \
--out models \
--fixed-model-name model-en