Warning: This document is for an old version of rasa NLU.

Improving your models from feedback

When the rasa_nlu server is running, it keeps track of all the predictions it’s made and saves these to a log file. By default log files are placed in logs/. The files in this directory contain one json object per line. You can fix any incorrect predictions and add them to your training set to improve your parser. After adding these to your training data, but before retraining your model, it is strongly recommended that you use the visualizer to spot any errors, see Visualizing training data.