Memoization Policy

Any one who can tell me the detail or fundamental of Memoization Policy. I read the RASA doc,but it just a little about Memoization Policy. And I also read rasa-core’s source code, but don’t understand. Who can help me? Thanks.

The Memoization Policy treats your stories not as ML training data, but as specific, literal instructions. That is, if the dialog so far matches a configured story exactly and the dialog engine needs to find the next action, the memoization policy will just use the next action in the story with confidence 100. If the running dialog does not match any configured story exactly, the Memoization policy will predict nothing and pass to the next policy in your policy ensemble. Normally this is one that uses machine learning and can deal with uncertainty like KerasPolicy.


Thanks a lot.