Hello, everyone. I got a question that, in the newest version of rasa, is there a way to train a core model based on a trained NLU model, like the version of tutorial weather chat bot shows. Because I think it’s more simple way to write story using interactive style with a corpus tagged and NLU model trained.
Hi, tyd, Thanks for your reply. Actually, I’ve tried interactive Learning, and found that it used to train rasa core as well as rasa nlu together, with untagged corpus. So i believed it’s time waste to tag again in interactive shell line under the situation that i’ve obtained large corpus tagged with intent and entity saved in json format. I’ve change the rasa version to satisfy my problem, Thanks.
@AlexRainHao I am curious to know more about this. What version of Rasa did you change to? What does it allow you to do that the current version does not?
sorry for replaying lately. In latest version rasa, we can conduct the training NLU and core in CMD Line with simple command，such as rasa train core, rasa interactive, that save time for python code if necessary. I found that, 1. For command ‘rasa train’, it guide you to train rasa NLU and Core together conveniently. 2. Whlie command ‘rasa train core’ or ‘rasa train nlu’ let us to train the two models separately. 3. The command rasa interactive developed for training NLU and core together in CMD line. The key problem is that all the 1,2,3 way to train core can not based on a certain designed NLU. Especially in interactive method, we need to input a conversation and tagged the intent, entity, then determine the next action, one after another. Therefore, It may be time wasting compared with the situation when we have tagged file which save by format either md or json. I change the rasa to the version rasa-nlu 0.12.3 and rasa-core 0.9.8, and python 3.6, then with the python code, i can train a nlu model, and use the model to train core in interactive way, which defined as train_online in rasa source code. So in this way, I can write lots of stories with the immediate prediction in the core model that optimized gradually. I’m a beginner in rasa, so this my person thought hope for helping you.