search

uber-archive / plato-research-dialogue-system

This is the Plato Research Dialogue System, a flexible platform for developing conversational AI agents.
Apache License 2.0
979 stars 191 forks source link

why need DM and NLG when train NLU? #12

Closed guhuawuli closed 4 years ago

guhuawuli commented 4 years ago

In tutorial (https://github.com/uber-research/plato-research-dialogue-system) I fond one part which is " Train NLU"

Since it is trained by

ludwig experiment \ --model_definition_file plato/example/config/ludwig/ludwig_nlu_train.yaml \ --data_csv data/DSTC2_NLU_sys.csv \ --output_directory models/camrest_nlu/sys/ why we still need: plato run --config CamRest_model_nlu.yaml

From CamRest_model_nlu.yaml I can see part DM and part NLG. Since it just need to train nlu what's the purpose of the other two part? 屏幕快照 2019-11-21 下午8 43 47

apapangelis commented 4 years ago

Hi, the purpose of the second part is to actually run a conversational agent that will load and use the trained NLU (that was trained with ludwig in the previous step). So, this agent still needs a DM and NLG to operate but if you see the NLU section it loads the trained model.

I hope this helps!