My current setup includes a bot that I had developed locally inside a Docker container and after being confident that it’s working well enough, I deployed it to an environment and connected that deployed bot to Rasa X so I could track the conversations coming in.
Here’s where I struggle though - When I send a message to the deployed bot, it identifies the intents and all the entities correctly. It is also able to store the identified entities as slots and continue the conversation, as defined. However, when I track that conversation in Rasa X, I see the bot capture it a bit differently.
In the Conversations tab, I see the conversation recorded as it proceeded in the deployed bot. But in the NLU inbox tab, I noticed that the utterance did not have all the right entities identified (there are two entities, but only one entity is marked).
I even tried the Talk to your bot tab, and it also did not identify the second entity. My understanding was that this is the same model. I’m not following why the bot in Rasa X is behaving so differently than the deployed bot. I was able to consistently reproduce this disparity for multiple examples.
A second issue with Rasa X’s Talk to the bot mode, is that when switching to interactive mode, I can’t annotate any entities; so I can’t correct any misidentified entities. A third issue is that the dropdown for selecting the right intent overflows down and gets clipped.
I’m currently using -
- Rasa X - 0.28.1
- rasa and rasa-sdk - 1.7.0
- OS - mac OS Catalina 10.15.6
- Chrome - Version 84.0.4147.89
Am I missing something? Many thanks for your help.