I’m working on a small bot in the Shell. I’m a little confused about how the “rasa train” command works and how I often I need to use it. Do i need to run rasa train in the shell every time I make any change to the code (the domain, story, or actions)? I’m running it on my local machine with rasa interactive right now but will later port it to a remote instance. Sometimes I may fix a spelling error or I may add an example utterance to the domain–pretty simple changes. It takes a bit to train every time I run the command and this makes the development pretty slow. I’d be happy to read the documentation about this (I couldn’t find it in the docs or on the forums on requirements for training the bot). Thanks!

Hi @travelmail26 Any changes to the nlu & domain sections require a retraining of the model to adjust with the changes. One thing you can do to minimize the training time is that you can try lowering the epochs in the configuration file. This might have negative impact on the accuracy. So you need to experiment and identify the right epochs for your dataset.

Do I need to train after each change?

Rasa Open Source

siriusraja (Raja Duraisingam) January 26, 2021, 5:36am 3

Hi @travelmail26

I would also suggest to have a look at a new feature for incremental training (can be used in certain cases) in version 2.2.0

3 Likes

Topic		Replies	Views
When do I retrain? Rasa Open Source	1	876	February 13, 2022
Rasa train Rasa Open Source	2	257	August 3, 2020
Incremental Training [Experimental] Important Updates	7	2688	July 19, 2022
Rasa Train model Rasa Open Source	3	292	July 1, 2021
Do you always have to run rasa train when making changes to domain.yml for response text? Rasa Open Source	3	1299	October 14, 2021

Do I need to train after each change?

Related Topics