Changing Config File

Rasa Open Source
1

Hi guys,

Do i need to re-train the model every time i change the configuration in config.yml file? I use the config file for training, but is it possible to change the config anytime then affecting the rasa-core and rasa-nlu? Using rasa core 0.14.4


user_policies = [KerasPolicy(), MemoizationPolicy(max_history=5),FallbackPolicy(nlu_threshold=0.8), FormPolicy(), MappingPolicy()]
nlu_interpreter = RasaNLUInterpreter('./models/nlu/global/latest')
action_endpoint = EndpointConfig(url=config['server']['actions_endpoint'])
nlg_endpoint = EndpointConfig(url=config['server']['nlg_endpoint'])
agent = Agent(policies=user_policies)
agent = agent.load('./models/dialogue', interpreter=nlu_interpreter, action_endpoint=action_endpoint,generator=nlg_endpoint)

Thanks

2

Hi,

Config.yml is holding your Rasa_NLU pipeline which define the architecture of the NLU model you are creating. Each time you change that config you need to train again to see changes.

Florian

3

Hi Florian,

So also do the same to rasa core?

Thanks for your help.

4

Rasa_Core is independent from Rasa_NLU so if you modify config.yml wich only affect Rasa_NLU you won’t have to train Rasa_Core again.

You need to train again Core only if you make changes in Core specific files like domain.yml or sotries.md from example.

Your welcome.

2 Likes
5

Gotcha, thanks for your help!