Version: 3.x
rasa.nlu.classifiers._fallback_classifier
FallbackClassifier Objects
class FallbackClassifier(IntentClassifier)
process
| process(message: Message, **kwargs: Any) -> None
Process an incoming message.
This is the components chance to process an incoming
message. The component can rely on
any context attribute to be present, that gets created
by a call to :meth:rasa.nlu.components.Component.create
of ANY component and
on any context attributes created by a call to
:meth:rasa.nlu.components.Component.process
of components previous to this one.
Arguments:
message
- The :class:rasa.shared.nlu.training_data.message.Message
to process.
is_fallback_classifier_prediction
is_fallback_classifier_prediction(prediction: Dict[Text, Any]) -> bool
Checks if the intent was predicted by the FallbackClassifier
.
Arguments:
prediction
- The prediction of the NLU model.
Returns:
True
if the top classified intent was the fallback intent.
undo_fallback_prediction
undo_fallback_prediction(prediction: Dict[Text, Any]) -> Dict[Text, Any]
Undo the prediction of the fallback intent.
Arguments:
prediction
- The prediction of the NLU model.
Returns:
The prediction as if the FallbackClassifier
wasn't present in the pipeline.
If the fallback intent is the only intent, return the prediction as it was
provided.