Hi everyone. Can anyone suggest me configuration of the config.yml file, Here is the config.yml file I am using which is giving good accuracy compared to the 2nd one which is very surprising considering DIET classifier is supposed to be better than logisticregression classifier.
Here are my 2 config files: this is config1.yml which I am currently using language: en recipe: default.v1
pipeline: ** - name: WhitespaceTokenizer** ** - name: RegexFeaturizer** ** - name: custom_components.tfidf_featurizer.TfIdfFeaturizer** ** analyzer: word** ** stop_words: english** ** min_df: 0.05** ** max_df: 1** ** sublinear_tf: true** ** norm: l2** ** lowercase: true**
** - name: LogisticRegressionClassifier** ** max_iter: 200** ** solver: lbfgs** ** tol: 0.0001** ** random_state: 42** ** ranking_length: 3**
** - name: CRFEntityExtractor**
** - name: EntitySynonymMapper**
** - name: FallbackClassifier** ** threshold: 0.7** ** ambiguity_threshold: 0.1**
# Configuration for 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 Policies for more information.
** - name: MemoizationPolicy** ** max_history: 2**
** - name: TEDPolicy** ** max_history: 5** ** epochs: 100** ** constrain_similarities: true**
** - name: RulePolicy** ** core_fallback_threshold: 0.4** ** core_fallback_action_name: “my_Api”** ** # enable_fallback_prediction: True**
This is the config2.yml which should be giving better results than the config1.yml
language: en recipe: default.v1
pipeline: ** - name: WhitespaceTokenizer** ** - name: RegexFeaturizer** ** - name: custom_components.tfidf_featurizer.TfIdfFeaturizer** ** analyzer: word** ** stop_words: english** ** min_df: 0.05** ** max_df: 1** ** sublinear_tf: true** ** norm: l2** ** lowercase: true**
** - name: DIETClassifier** ** epochs: 100** ** constrain_similarities: true**
** - name: EntitySynonymMapper**
** - name: FallbackClassifier** ** threshold: 0.7** ** ambiguity_threshold: 0.1**
# Configuration for 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 Policies for more information.
** - name: MemoizationPolicy** ** max_history: 2**
** - name: TEDPolicy** ** max_history: 5** ** epochs: 100** ** constrain_similarities: true**
** - name: RulePolicy** ** core_fallback_threshold: 0.4** ** core_fallback_action_name: “my_Api”** ** # enable_fallback_prediction: True**
The config file with DIET classifier is giving poor results compared to the file with logistic regression classifier. Can config1.yml be improved in any way if my goal is just improving the accuracy of 1. Intent detection and 2. Entity extraction. ?