You can create and manage your training pipeline configurations in the « Configuration » section.
Click on the « New configuration » button to create a new configuration. Give it a name and upload your configuration file (.yml). Use the « note section » to describe the main characteristics of this configuration. Click on the « Save » button to finalize the creation of your configuration.
Create a configuration file
In order to train your model, you will need to provide it with a configuration file.
The goal of this file is to describe the training pipeline (i.e, the list and sequence of Machine Learning components) that will be used to process your data and train your model. Four main categories of components can be used:
- Intent featurizers
- Intent classifiers
- NER (Named-entity recognition)
Each component can be provided by multiple « framework » families such as SPACY, MITIE, JIEBA, … Some « families » can mix their components with each others, some can’t. For example you can mix an « intent_featurizer_spacy » with a « ner_duckling_http » but can’t mix it with a « ner_mitie ».
The documentation of all the components available in RASA NLU is available on the following link : https://rasa.com/docs/nlu/master/components/
The Yaml file has to be formatted as shown in the example below:
--- language: "fr" pipeline: - name: "nlp_spacy" model: "fr" - name: "tokenizer_spacy" - name: "intent_featurizer_spacy" - name: "ner_crf" - name: "ner_synonyms" - name: "intent_classifier_sklearn"
Supported langages and components
The platforms natively supports English and French training components.
German, spanish, portuguese, italian, dutch and chinese can be supported on demand. Please contact us at email@example.com if you need one of these languages.
MITIE components have been deactivated by default in the platform, due to very high computing times. Contact us at firstname.lastname@example.org if you need to activate it for a specific project.
Lookup and Regex components are not supported yet by the platform.