Best Config.yml

Rasa Open Source
1

hi, i have some doubts about config.yml file, my faqbot is confusing itself

My stories are formatted as a single story and the longest story has 3 intents. i’m using forms, slots and custom actions. It has 557 intents and 4719 intents examples.

i’t’s my config.yml file:

language: pt

pipeline:

  • name: “SpacyNLP”
  • name: “SpacyTokenizer”
  • name: “RegexFeaturizer”
  • name: “CRFEntityExtractor”
  • name: “EntitySynonymMapper”
  • name: “CountVectorsFeaturizer” strip_accents: “unicode” analyzer: “char_wb” min_ngram: 1 max_ngram: 4
  • name: “CustomLemma”
  • name: “EmbeddingIntentClassifier”

policies:

  • name: FormPolicy
  • name: AugmentedMemoizationPolicy max_history: 1
  • name: KerasPolicy validation_split: 0.2 batch_size: 200 epochs: 800
  • name: MappingPolicy
  • name: FallbackPolicy nlu_threshold: 0.7 core_threshold: 0.7 fallback_action_name: “utter_default”

CustomLemma is my custom component. I’m training the chatbot with augmentation parameter “–augmentation 300”

Is there a better way to config my config.yml ?

2

@KPrado could you give an example of what kind of situations the bot is failing?

3

Hi @akelad, I have an intent “saudacao”

image

It’s registered in the domain.yml and stories.md file, but when we ask a question about this intent, the accuracy is low Ex:

image
image1357×149 57.1 KB

image
image1406×98 35.1 KB

image
image1409×63 22.1 KB

image
image1893×464 67.2 KB

Do you know what may be happening?

4

You have a lot of intents examples. You could use the supervised embeddings pipeline. This pipeline will create its own vectors. This way the model will be better adapted to your data and may provide better values. For this just replace the first two components of your pipeline with the Whitespace Tokenizer component.