config.yml

language: zh
pipeline:
- name: HFTransformersNLP
  model_weights: "bert-base-chinese"
  model_name: "bert"

- name: LanguageModelTokenizer
- name: LanguageModelFeaturizer

- name: CountVectorsFeaturizer
- name: CountVectorsFeaturizer
  analyzer: char_wb
  min_ngram: 1
  max_ngram: 4

- name: DIETClassifier
  epochs: 30
  learning_rate: 0.005
  num_transformer_layers: 0
  embedding_dimension: 30
  weight_sparcity: 0.8
  hidden_layer_sizes:
    text: [256, 128]

policies:
  - name: AugmentedMemoizationPolicy
  
  - name: TEDPolicy
    epochs: 100
    featurizer:
    - name: MaxHistoryTrackerFeaturizer
      max_history: 5
      state_featurizer:
        - name: BinarySingleStateFeaturizer


  - name: FallbackPolicy
    fallback_action_name: 'action_default_fallback'
    nlu_threshold: 0.5
    core_threshold: 0.3
    ambiguity_threshold: 0.1
  
  - name: MappingPolicy