Response Selectors high loss but acc is 100 %!

Hi guys I am working on building a digital assistant using rasa and I am facing this problem

  1. While training the response selectors have a very high loss eg 5.6 but the r_acc (accuracy) is 100 . why is this happening ?

Well the intents are classified properly but I want to know if this is a common thing or will it end up creating problems in the future.

Here is my config.yml file

language: en

pipeline:
# No configuration for the NLU pipeline was provided. The following default pipeline was used to train your model.
# If you'd like to customize it, uncomment and adjust the pipeline.
# See https://rasa.com/docs/rasa/tuning-your-model for more information.
  - name: WhitespaceTokenizer
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: "CountVectorsFeaturizer"
    # Analyzer to use, either 'word', 'char', or 'char_wb'
    "analyzer": "word"
    # Set the lower and upper boundaries for the n-grams
    "min_ngram": 1
    "max_ngram": 1
    # Set the out-of-vocabulary token
    "OOV_token": "_oov_"
    # Whether to use a shared vocab
    "use_shared_vocab": False
  - name: LanguageModelFeaturizer
    # Name of the language model to use
    model_name: "bert"
    # Pre-Trained weights to be loaded
    model_weights: "rasa/LaBSE"
    cache_dir: null
  - name: RegexEntityExtractor
    case_sensitive: False
    use_lookup_tables: True 
  - name: "DucklingEntityExtractor"
    # url of the running duckling server
    url: "http://localhost:8000"
    # dimensions to extract
    dimensions: ["time", "amount-of-money", "distance","email","phone-number"]
    timeout : 3
  - name: DIETClassifier
    epochs: 150
    constrain_similarities: true
    # model_confidence: linear_norm
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
    retrieval_intent: faq
  - name: ResponseSelector
    epochs: 100
    retrieval_intent: chitchat
  - name: ResponseSelector
    epochs: 100
    retrieval_intent: inform
  - name: FallbackClassifier
    threshold: 0.55
    ambiguity_threshold: 0.05

# Configuration for Rasa Core.
# https://rasa.com/docs/rasa/core/policies/
policies:
# # No configuration for policies was provided. The following default policies were used to train your model.
# # If you'd like to customize them, uncomment and adjust the policies.
# # See https://rasa.com/docs/rasa/policies for more information.
  - name: RulePolicy
  - name: MemoizationPolicy
    max_history: 3
  - name: TEDPolicy
    max_history: 5
    epochs: 300
    constrain_similarities: true

And here is the nlu.yml file containing the response selector utterances I had to move it here instead of domain.yml as it becz of the user warnings

responses:

  utter_faq/contact_inquiry:
  - text: I am calling you because of the request you have made for insurance quote are you interested ?
  - text: The reason I am calling Is because of the request you had submitted online are you interested in looking for a better policy ?
  - text: We have received a request for you for a auto insurance quote are you still looking ?
  - text: I am calling you in response to the request you had submitted online for a auto quote are you still looking ?
  utter_faq/personal_questions:
  - text: My name is Reha I am a independent financial agent calling you to get a better quote and saving money are you interested ?
  - text: I am Reha hi I work as a independent financial agent calling you to get a better quote and saving money are you interested ?
  - text: My name is Reha I am a independent financial agent I work for <company_name> are you interested in looking for a quote ?
  - text: Reha,I am a independent financial agent working for <company_name> I can help you get a better quote are you interested ?
  utter_faq/email:
  - text: yes I will mail you the details should
  - text: sure will mail you the all the policy details
  - text: ya will mail you the details
  - text: ya will email you the detials.
  utter_faq/website:
  - text: our website is <website link>
  utter_faq/ask_question:
  - text: Yes please go ahead
  - text: yes sure you can
  - text: yes go ahead
  - text: dont hesitate just ask anything about the insurance
  - text: yes sure please go ahead
  - text: sure go on
  utter_faq/policy_questions:
  - text: <respond to policy query>
  utter_faq/price_questions:
  - text: <utter_price_response>
  - text: yes the <utter_price_response>
  - text: yes its is the <utter_price_response>
  - text: <utter_price_response>
  utter_chitchat/ask_howdoing:
  - text: I am fine thanks for asking
  - text: I am going great thank for asking
  - text: fine just doing my work thanks for asking
  - text: great thanks for asking
  - text: ok thanks for asking
  utter_chitchat/ask_isbot:
  - text: I am a digital assistant powered by ai
  - text: Yes I'm a digital person powered by ai
  - text: Yes I am REHA a digital bot powered by Artificial intelligence
  - text: The names Reha a digital person to help you find the best quotes.
  utter_chitchat/ask_ishuman:
  - text: No I am not a human My name is Reha I am a Artificial intelligence lifeform
  - text: Not a human a digital assistant created to help you find the best insurance
  - text: I am a digital Assistant not a human
  utter_inform/customer_basic_information:
  - text: <customer_name> to get you the best policy I would like to ask you some details if that ok with you ?
  utter_inform/customer_auto_information:
  - text: <customer_name> to get you the best policy I would like to ask you some details if that ok with you ?
  utter_inform/current_insurer:
  - text: <customer_name> to get you the best policy I would like to ask you some details if that ok with you ?

Rasa version

  • rasa 2.6.0
  • rasa-sdk 2.6.0
  • rasa-x 0.0.1
  • rasalit 0.1.2

I am also including my domain.yml and nlu.yml file domain.yml (13.4 KB)

nlu.yml (31.8 KB)