rasa_contrib is a addon package for rasa. It provide some useful/powerful addition component.
Currently, it includes:
-
addons_intent_classifier_textcnn_tf
TextCNN based intent classifier, based on TensorFlow
-
addons_intent_classifier_textcnn_paddle
TextCNN based intent classifier, based on PaddlePaddle
-
addons_ner_bilstm_crf_tf
Embedding+BiLSTM+CRF based NER extractor, based on TensorFlow
-
addons_ner_bilstm_crf_paddle
Embedding+BiLSTM+CRF based NER extractor, based on PaddlePaddle
-
bert_text_featurizer get BERT-based text vector feature
-
bert_char_featurizer get BERT-based char/word vector feature
It also includes (but still work in progress):
-
MicroAddonsTokenizer
Chinese tokenizer component, based on MicroTokenizer
-
StackedBilstmTensorFlowPolicy
Stacked Bilstm based dialog policy, based on TensorFlow
-
StackedBilstmPaddlePolicy
Stacked Bilstm based dialog policy, based on PaddlePaddle
Using the class path to the place where you should given a component name in config.yaml. This is a feature of rasa, see here for more document from rasa official document.
For example, your config.yml can be:
language: "zh"
pipeline:
- name: "rasa_contrib.nlu..TensorflowNLP"
- name: "rasa_contrib.nlu..BilstmCrfTensorFlowEntityExtractor"
max_steps: 600
- name: "rasa_contrib.nlu.TextCnnTensorFlowClassifier"
max_steps: 600
policies:
- name: MemoizationPolicy
- name: rasa_contrib.core.StackedBilstmTensorFlowPolicy