Closed indam23 closed 2 years ago
The problem is solved. You can specify --domain
argument in CLI and you'll get an expected result in both 2.8.x
and 3.0.x
e.g. rasa test nlu --cross-validation --domain domain
Hello,
I have the same problem.
Rasa Version : 2.8.3 Minimum Compatible Version: 2.8.0 Rasa SDK Version : 2.8.1 Rasa X Version : None Python Version : 3.8.9 Operating System : Windows-10-10.0.22000-SP0
And here is the issue:
Hello,
I have the same problem.
Rasa Version : 2.8.3 Minimum Compatible Version: 2.8.0 Rasa SDK Version : 2.8.1 Rasa X Version : None Python Version : 3.8.9 Operating System : Windows-10-10.0.22000-SP0
And here is the issue:
![Uploading Στιγμιότυπο οθόνης 2022-05-20 114830.png…]()
@EvanMath Can you please upgrade to the latest version of 2.8 (2.8.27) and report if the problem persists?
@melindaloubser1 Thank you. it is solved.
Rasa Open Source version
2.8.16, however this would also be true of 3.0.0
What happened?
I have a directory structure with split domains, like this:
ResponseSelector responses are in
domain/domain_responses.yml
, and this works fine for training and prediciton.I am using ResponseSelector with
use_text_as_label: true
as e.g. this minimal config:When running NLU cross-validation, the responses are not available to ResponseSelector. Therefore, it falls back to using the response labels, and gives 0 confidence prediction for everything. This is different from the behaviour at training time, so it means my test results do not reflect real-life results.
Command / Request
Relevant log output
Expected type of errors in response_selection_errors.json:
Actual type of errors in response_selection_errors.json - no prediction at all:
Definition of done