Rest responses are different from "shell"

I am using the same conversation scenario in “rasa shell” and “rest” requests, but the responses are different.



Help please. Where should I look? What to check? Any thoughts?

hi @dmlapteacru ! hmm that does sound strange. Can you please run rasa with --debug switched on? hopefully that tells us what’s happening.

are you definitely having the exact same conversation (same messages in same order) in both cases?

@amn41 Thanks for responding, I have increased the epochs in KerasPolicy and changed the intents name to the unique ones like uuid (cause the only who needs to trigger them its me, not the user) and now there are the same results.

But I am running in a issue where I get the Fallback policy response, even when the confidence is 1.0 and filling the entity.



Settings: image

Any help on this please?

the debug log mentions that the keras policy predicted an action_listen right after a user message, which would mean that nothiing happens. To prevent that, the fallback action kicks in. So the problem is that the Keras policy is predicting action_listen with high confidence. Can you think of anything in your stories that might cause it to predict that?

@amn41 There is no data in that intent. It is just ##intent:save_token, I use it to save token for user, so I can use it for backend requests. Would be great you to look into what I have done, I would describe the purpose of the bot and I would listen the better approaches from you than I used :slight_smile: But I know you have a lot of work…You guys are doing a great job.