I have implemented the contextual interjections as described in Handling Unexpected Input
My config policies are as follows.
policies:
- name: RulePolicy
core_fallback_threshold: 0.7
core_fallback_action_name: "action_human_handoff"
enable_fallback_prediction: True
- name: AugmentedMemoizationPolicy
max_history: 2
- name: TEDPolicy
epochs: 300
epoch_fraction: 0.5
connection_density: 0.3
constrain_similarities: true
What happens is both AugmentedMemoizationPolicy and TEDPolicy predicts the response action for the interjection with confidence of 1.0 and 0.85 respectively.
Since there is a form already in loop, Rule Policy is predicting the form with confidence of 1.0 This happens because Rule policy has higher priority.
In such scenarios, how would these contextual interjections work ?
