The model I trained for first time works fine, but retraining the model breaks the conversation flow

Rasa Open Source
1

I have the following in my config.yml.

pipeline:

  - name: WhitespaceTokenizer
  - name: RegexEntityExtractor
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: char_wb
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 100
    constrain_similarities: true
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
    constrain_similarities: true

policies:
  - name: TEDPolicy
    max_history: 5
    epochs: 100
    constrain_similarities: true
    model_confidence: linear_norm
  - name: MemoizationPolicy
  - name: RulePolicy

The problem is, the stories and rules are created and the model is trained using ‘rasa train’ command. It works fine accordingly as expected. But when I retrain the model, the behavior of the rasa assistant changes and the flow breaks. Not sure why this is happening. Any help would be appreciated.

2

Machine learning is random :slight_smile:

If your outputs are not consistent, consider adding more data (intent examples and stories). Or, you can add random_seed: 1 (or any number) in DIETClassifier, ResponseSelector, and TEDPolicy. For example:

  - name: DIETClassifier
    epochs: 100
    constrain_similarities: true
    random_seed: 1
3

@actorkrishna Hi, can you delete older trained models and train again and run ? This approach is valid if you believe your code is 100% correct.

1 Like
4

It won’t train a new model if nothing is changed in the code. It tells me that I can use the latest model. The problem happened after I deleted the old model and retrained the rasabot

5

@actorkrishna Share the logs it will give a better understanding.