Skip to content


The SemanticMapFeaturizer is an experimental sparse featurizer developed by Rasa. It can only be used in combination with pre-trained embedding files, which you can find here. Please refer to our blog posts for more details.

Configurable Variables

  • pretrained_semantic_map: Path to downloaded/saved semantic map embeddings (the unpacked json file)
  • pooling: The pooling operation to use for the sentence features (sum (default), mean, or merge)

Basic Usage

The configuration file below demonstrates how you might use the semantic map embeddings. In this example we're building a pipeline for the English language and we're assuming that you've saved embeddings upfront as saved/beforehand/rasa-sme-wikipedia-en-64x64-v20201120.json.

language: en

  - name: WhitespaceTokenizer
  - name: LexicalSyntacticFeaturizer
  - name: rasa_nlu_examples.featurizers.sparse.SemanticMapFeaturizer
    pretrained_semantic_map: "saved/beforehand/rasa-sme-wikipedia64x64-en-v20201120.json"
  - name: DIETClassifier
    epochs: 100