My rasa model make poor predictions when i add certain part in the story

I’m working on a virtual assistant that check an agent disponability, call them , check an appointment and send emails. after training ,it always respond with action_default_fallback i tried certain components in the config.yml :

    pipeline:
  - name: SpacyNLP
    model: "fr_core_news_sm"
  - name: SpacyTokenizer
  - name: SpacyFeaturizer
    pooling: mean
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: "char_wb"
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 200
  - name: EntitySynonymMapper
  - name: RegexEntityExtractor
  - name: ResponseSelector
    epochs: 200
    model_confidence: linear_norm
    constrain_similarities: True

policies:
  - name: AugmentedMemoizationPolicy
    max_history: 4
  - name: TEDPolicy
    max_history: 5
    epochs: 150
  - name: RulePolicy
    core_fallback_threshold: 0.4
    core_fallback_action_name: "action_default_fallback"
    enable_fallback_prediction: True

the stories:

version: "2.0"
stories:
- story: greeting
  steps:
  - intent: greet
  - action: utter_greet
  - action: utter_help
  - checkpoint: check_greeting_and_help

- story: say goodbye
  steps:
  - checkpoint: check_greeting_and_help
  - checkpoint: goodbye
  - intent: goodbye
  - action: utter_goodbye

- story: say thank you
  steps:
  - checkpoint: goodbye
  - checkpoint: check_thank_you
  - intent: thank_you
  - action: utter_thanks

- story: demand agent by name + service available
  steps:
  - checkpoint: check_greeting_and_help
  - intent: informAgentService
    entities:
    - agent_name: "wassim"
    - service: "technique"
  - slot_was_set:
    - agent_name: "wassim"
    - service: "technique"
  - action: utter_intent_reach_agent
  - action: utter_patienter
  - action: action_reach_agent
  - checkpoint: check_agent_by_name_call_confirmation

- story: demand agent by name available
  steps:
  - checkpoint: check_greeting_and_help
  - intent: informAgent
    entities:
    - agent_name: "wassim"
  - slot_was_set:
    - agent_name: "wassim"
  - action: utter_ask_service
  - intent: infromserviceafteragent
    entities:
    - service: "technique"
  - slot_was_set:
    - service: "technique"
  - action: utter_intent_reach_agent
  - action: utter_patienter
  - action: action_reach_agent
  - checkpoint: check_agent_by_name_call_confirmation

- story: avaiblable agent demand for call confirmation answer oui
  steps:
  - checkpoint: check_agent_by_name_call_confirmation
  - action: utter_confirmation
  - intent: affirm
  - action: utter_patienter
  - action: action_call_agent
  - checkpoint: check_thank_you

- story: avaiblable agent demand for call confirmation answer non
  steps:
  - checkpoint: check_agent_by_name_call_confirmation
  - action: utter_confirmation
  - intent: deny
  - action: utter_other_proposition
  - checkpoint: check_asked_other_proposition

- story: demand agent by service available S0
  steps:
  - checkpoint: check_greeting_and_help
  - intent: informService
    entities:
    - service: "technique"
  - slot_was_set:
    - service: "technique"
  - action: utter_intent_reach_agent
  - action: utter_patienter
  - checkpoint: check_call_confirmation


- story: demand agent by service available s1.0
  steps:
  - checkpoint: check_call_confirmation
  - checkpoint: chosen_proposition_agent
  - action: action_reach_agent_by_service
  - action: utter_confirmation
  - intent: affirm
  - action: utter_patienter
  - action: action_call_agent
  - checkpoint: check_thank_you

- story: demand agent by service available s1.1
  steps:
  - checkpoint: check_call_confirmation
  - slot_was_set:
    - agent_name: null
  - checkpoint: chosen_proposition_agent
  - action: action_reach_agent_by_service
  - action: utter_confirmation
  - intent: deny
  - action: utter_other_proposition
  - checkpoint: check_asked_other_proposition

- story: demand agent by service available s2
  steps:
  - checkpoint: chosen_proposition_agent
  - action: action_reach_agent_by_service
  - checkpoint: check_asked_other_proposition

- story: demand agent by name + service not available
  steps:
  - checkpoint: check_greeting_and_help
  - intent: informAgentService
    entities:
    - agent_name: "franck"
    - service: "comptabilité"
  - slot_was_set:
    - agent_name: "franck"
    - service: "comptabilité"
  - action: utter_intent_reach_agent 
  - action: utter_patienter
  - action: action_reach_agent   
  - checkpoint: check_asked_other_proposition

- story: demand agent by name not available
  steps:
  - checkpoint: check_greeting_and_help
  - intent: informAgent
    entities:
    - agent_name: "franck"
  - slot_was_set:
    - agent_name: "franck"
  - action: utter_ask_service
  - intent: infromserviceafteragent
    entities:
    - service: "comptabilité"
  - slot_was_set:
    - service: "comptabilité"
  - action: utter_intent_reach_agent
  - action: utter_patienter
  - action: action_reach_agent   
  - checkpoint: check_asked_other_proposition

- story: choose other proposition message
  steps:
  - checkpoint: check_asked_other_proposition
  - intent: choose_message
  - action: utter_write_send_message
  - action: utter_write_send_message_button
  - or:
    - intent: text
    - intent: vocal
  - action: utter_write
  - checkpoint: goodbye

- story: choose other proposition autre agent
  steps:
  - checkpoint: check_asked_other_proposition
  - intent: choose_autre_agent
  - checkpoint: chosen_proposition_agent

- story: choose other proposition rendez vous
  steps:
  - checkpoint: check_asked_other_proposition
  - intent: choose_rendezvous
  - action: utter_ask_for_date
  - intent: informTime
    entities:
    - date: "2020-02-30"
    - time: "15:30"
  - slot_was_set:
    - date: "2020-02-30"
    - time: "15:30"
  - action: action_set_rendez_vous
  - checkpoint: goodbye
  - checkpoint: check_thank_you

so i tried to eliminate the check appointment part from the stories and it works fine , whenever i put it again it returns action_default_fallback

Could you post your stories?

added

One thing I noticed is that you have a lot of stories that could be changed into rules. Combining rules and stories is the best way to build a conversational model.

In your case, stories such as:

  steps:
  - checkpoint: goodbye
  - checkpoint: check_thank_you
  - intent: thank_you
  - action: utter_thanks

Can be changed into rules and it is certainly a good way to improve your model. Regarding your issue, I will review your data in a bit and get back to you.