Evaluation with multi intent classification

When I run the rasa_nlu.evaluate module it seems like it treats my “intent1+intent2” as single intents. May be I am not defining the multi-intents incorrectly in the data? When it is training the model it also complains that there are a few examples for the intent “intent1+intent2”, so I think it treats the multi intent that I am marking down in the data as a single intent.

When I get the errors through setting --errors parameter it includes things like intent: “intent1+intent2” predicted: “intent1”.

Can someone help me understand whether evaluation just does not support multi-intent, or I am using this feature incorrectly.

Never tried multiple intent, but I know it only works with the tensorflow pipeline, are you using it or are you using the spacy pipeline ?

See : How to handle multiple intents per input using Rasa NLU TensorFlow pipeline

I am using the tensorflow pipeline