What if i want to train NLU alone in RASA 1.0?

Rasa Open Source
1

Hi all,

I have just started with RASA 1.0 and i would like to know how to train the nlu model alone with /model/train API.

As per the documentations payload of the /model/train API is

{ "domain": “intents:\n - greet\n - goodbye\n - affirm\n - deny\n - mood_great\n - mood_unhappy\n\nactions:\n - utter_greet\n - utter_cheer_up\n - utter_did_that_help\n - utter_happy\n - utter_goodbye\n\ntemplates:\n utter_greet:\n - text: “Hey! How are you?”\n\n utter_cheer_up:\n - text: “Here is something to cheer you up:”\n image: “https://i.imgur.com/nGF1K8f.jpg”\n\n utter_did_that_help:\n - text: “Did that help you?”\n\n utter_happy:\n - text: “Great carry on!”\n\n utter_goodbye:\n - text: “Bye””,

"config": “language: en\npipeline: supervised_embeddings\npolicies:\n - name: MemoizationPolicy\n - name: KerasPolicy”,

"nlu": “## intent:greet\n- hey\n- hello\n- hi\n## intent:goodbye\n- bye\n- goodbye\n- have a nice day\n- see you\n## intent:affirm\n- yes\n- indeed\n## intent:deny\n- no\n- never\n## intent:mood_great\n- perfect\n- very good\n- great\n## intent:mood_unhappy\n- sad\n- not good\n- unhappy”,

"stories": “## happy path\n* greet\n\n - utter_greet\n\n* mood_great\n\n - utter_happy\n\n## sad path 1\n* greet\n\n - utter_greet\n\n* mood_unhappy\n\n - utter_cheer_up\n\n - utter_did_that_help\n\n* affirm\n\n - utter_happy\n\n## sad path 2\n* greet\n\n - utter_greet\n\n* mood_unhappy\n\n - utter_cheer_up\n\n - utter_did_that_help\n\n* deny\n\n - utter_goodbye\n\n## say goodbye\n* goodbye\n\n - utter_goodbye”,

"force": false,

"save_to_default_model_directory": true }

My training set and stories are so huge that i cannot afford to make both NLU and core training every time i make some minor changes in the NLU.So is there any way to make only nlu training in rasa 1.0.

Regards

2

Rasa will automatically skip training any parts where it doesn’t need to. So if your stories and policies haven’t changed, it’ll skip training the core model

3

Hi @amn41

I was able to figure that out when i tried to retrain the model, great feature from RASA kudos to that :slight_smile: .Yet my other constraint is with nlu_data format.I guess the API will only take md file as the nlu payload ??

Is there any way i could take json file as the nlu payload? Since i have already built a utility which could convert my csv file to json format, so i cannot afford to waste my time for converting the json back to md.

Regards

4

you can convert between md and json with rasa data convert

5

The Core keeps on retraining even if there are no changes in stories or config.

When i checked the fingerprint.json i could find out the stories checksum has been changed with every model. I am currently training using the same nlu,config and stories …etc.

Any way to solve this ??

6

if you haven’t changed anything but the model is retraining anyway, then that’s a bug. Is it the same as this issue? Sometimes NLU requires retraining when domain file is changed · Issue #4691 · RasaHQ/rasa · GitHub If not then please create a new issue with steps to reproduce the bug