-
**Description of Problem**: Currently, all ML components can be configured to use either `margin` loss or `cross_entropy`(formerly known as `softmax`) loss. Results have shown that `cross_entropy` los…
-
**Description of Problem**:
RegexFeaturizer is using a single CPU core, causing slow training instead of leveraging idle CPU cores. In our case, it is taking ~1 hour, which can easily become ~5 mins.…
-
**Rasa version**: 1.10.10
**Python version**: 3.7
**Operating system** (windows, osx, ...): Ubuntu
**Issue**: Adding word vectors in a layer before classifier is problematic in terms of memory. Wor…
-
**Description of Problem**: When training a Rasa pipeline on a large dataset, you don't know how many epochs you need to train the ML component for. However, as the training proceeds, monitoring train…
-
### Rasa Open Source version
2.8.14
### Rasa SDK version
_No response_
### Rasa X version
_No response_
### Python version
3.8
### What operating system are you using?
Linux
### What hap…
-
[IBM Watson Assistant] ([https://www.ibm.com/watson/ai-assistant/](https://www.ibm.com/watson/ai-assistant/)) is a popular solution to build ChatBots. We do have import from multiple systems - DialogF…
-
**Description of Problem**:
LexicalSyntacticFeaturizer computes unnecessary operations when creating features.
[https://github.com/RasaHQ/rasa/blob/master/rasa/nlu/featurizers/sparse_featurizer/lexi…
-
### What problem are you trying to solve?
I was a user of Rasa back in 2018 and early 2019\. One of the major issues that I faced when building a chatbot was the lack of data that I had for training …
-
### Rasa Open Source version
2.8.2
### Rasa SDK version
_No response_
### Rasa X version
_No response_
### Python version
3.7
### What operating system are you using?
Linux
### What happ…
-
**Description of Problem**: In Rasa 2.x, we can now add `metadata` to intents for custom NLU pipeline components. It would be great to have this flexibility for other NLU data such as synonyms.
I am …