Hi there, I just found the multiProjectsImporter feature, which would be perfect for our Bot. Currently we have different parts of functionality handling different subjects in our Bot. When a user starts a conversation he can choose which functionality he would like to use. At the moment this is one big Rasa project and our training takes ages (90 min!) and every time we change something and want to test our changes we need to retrain obviously. Therefore it would be great to split up the project. Is there a way to train a sub-project seperately? Meaning could you train each part individually and than merge the resulting training-files somehow? This would reduce our time effort immensely cause a change in one part of the Bot would not result in the need to retrain the whole project.
Therefore it would be great to split up the project. Is there a way to train a sub-project seperately?
You can train and speak to the model of a subproject separately, but you can’t merge the trained models. While (I think) this could theoretically be done for rule based policies such as MemoizationPolicy it wouldn’t work for any component / policy which involves machine learning, One of the problems would for example be that the input / output dimensions of each subproject’s model would be different.
Thanks Tobias, too bad that this is not possible, but it makes total sense. It’s still useful to test each feature separately and keep the project structure tidy. But I think the Documentation is a bit confusing on that particular point: Rasa docu on multi project importer ‘During the training process Rasa will import all required training files, combine them, and train a unified AI assistant. The merging of the training data happens during runtime, so no additional files with training data are created or visible.’