Hi, I’m wondering what the difference is between training in rasa cli using rasa train and training in the GUI with Rasa X?
Are those supposed to be the same under the hood, or is there an intentional difference?
Hi @jabberlope, the training should be the same under the hood. The training data for the training in Rasa X comes from your training data tab and the stories in your stories tab. Are you noticing differences or having issues?
@erohmensing, thanks for the reply! The reason for my question was I noticed that there is no console output for the Keras training layers when training via RASA X, so I wanted to confirm my assumption that the two training methods were identical.
Perhaps this output is suppressed when not running rasa x in debug mode?
Yes, I believe we wanted to hide that output in the rasa X logs so that they didn’t take up so much space, as ideally one wouldn’t be looking at the logs much except to find out what happened if something went wrong.
UPDATE: I have found that when I introduce formatting errors into the markdown that the rasa train script will complain, but when I retrain in rasa x those errors are being suppressed.
Case in point, I installed an MD formatter and it broke the NLU.md by placing a space after every ## intent:intent_name_here and I only discovered this after running rasa train because rasa x would just fail and say there was nothing to update.
That sounds like a bug then, those errors should definitely not be surpressed. Would you mind creating a bug report in a github issue at github.com/rasahq/rasa, and also include the full stack trace of how both rasa train and rasa x fail?
Glad to. Will take a bit to put an issue together, but I’ll get something in within the week.
Sure thing. Appreciate the effort!