Hi,
I have structured the NLU data in my data folder by having an individual file for every intent, which works fine for training. Unfortunately, when using Rasa X (guess its the interactive mode) newly classified text examples get written to random files. So that I end up with files that contain multiple intents (and often only a single example). Guess this shouldnt be a problem for training, however it is annoying for organizing my data. Is this expected behaviour or am I doing something wrong?
This is probably a general problem of the interactive mode, but at least in the CLI its asking for a (single) file to write to.
the issue is you did not choose the correct config file you must choose the correct pipeline to have the correct the training you can use space to detect names but I do not have the full idea I guess that page will help you
I dont really understand your reply, but I think its not an answer to my question. The problem is that Rasa automatically writes intents that I classify manually to random files. So for example I have a file ‘affirm.md’ with the intent ‘affirm’ and values ‘yes, yep, absolutely’. Now I use interactive mode and write ‘correct’ to affirm and manually label it as ‘affirm’ intent, because rasa doesnt know it yet. Now its seems like its totally random in which file the newly labelled ‘correct’ ends up. Might end up in deny.md, which then holds the intents ‘deny’ and ‘affirm’. Doesnt seem to hurt any functionality, but is annoying.
Hi @IgNoRaNt23, Rasa X sorts new NLU examples into the oldest existing file. In future, we may improve that by sorting new intent examples into the file in which that intent is already present. For new intents, the assignment will probably still be based on something like the oldest file, or the file with the largest number of examples