Migration Guide

This page contains information about changes between major versions and how you can migrate from one version to another.

0.14.x to 0.15.0

Warning

This is a release breaking backwards compatibility. Unfortunately, it is not possible to load previously trained models (as the stored file names have changed as well as the configuration and metadata). Please make sure to retrain a model before trying to use it with this improved version.

model configuration

• The standard pipelines have been renamed. spacy_sklearn is now pretrained_embeddings_spacy and tensorflow_embedding is now supervised_embeddings.
• Components names used for nlu config have been changed. Use component class name in nlu config file.

custom components

• The signature of Component’s methods have been changed:
  • load(...), create(...) and cache_key(...) methods additionally take component’s meta/config dicts.
  • persist(...) method additionally takes file name prefix. Change your custom components accordingly.

function names

• rasa_nlu.evaluate was renamed to rasa_nlu.test
• rasa_nlu.test.run_cv_evaluation was renamed to rasa_nlu.test.cross_validate
• rasa_nlu.train.do_train() was renamed to to rasa_nlu.train.train()

0.13.x to 0.14.0

• /config endpoint removed, when training a new model, the user should always post the configuration as part of the request instead of relying on the servers config.
• ner_duckling support has been removed. Use DucklingHTTPExtractor instead. More info about DucklingHTTPExtractor can be found at Duckling HTTP Extractor.

0.13.x to 0.13.3

• rasa_nlu.server has to be supplied with a yml file defining the model endpoint from which to retrieve training data. The file location has been passed with the --endpoints argument.

Example

model:
    url: http://my_model_server.com/models/default/nlu/tags/latest
    token: my_model_server_token

0.12.x to 0.13.0

Warning

This is a release breaking backwards compatibility. Unfortunately, it is not possible to load previously trained models as the parameters for the tensorflow and CRF models changed.

CRF model configuration

The feature names for the features of the entity CRF have changed:

Please change these keys in your pipeline configuration of the CRFEntityExtractor components features attribute if you use them.

0.11.x to 0.12.0

Warning

This is a release breaking backwards compatibility. Unfortunately, it is not possible to load previously trained models (as the stored file formats have changed as well as the configuration and metadata). Please make sure to retrain a model before trying to use it with this improved version.

model configuration

We have split the configuration in a model configuration and parameters used to configure the server, train, and evaluate scripts. The configuration file now only contains the pipeline as well as the language parameters. Example:

language: "en"
pipeline:
- name: "SpacyNLP"
  model: "en"               # parameter of the spacy component
- name: "EntitySynonymMapper"

persistors:

• renamed AWS_REGION to AWS_DEFAULT_REGION
• always make sure to specify the bucket using env BUCKET_NAME
• are now configured solely over environment variables.

0.9.x to 0.10.0

• We introduced a new concept called a project. You can have multiple versions of a model trained for a project. E.g. you can train an initial model and add more training data and retrain that project. This will result in a new model version for the same project.

0.8.x to 0.9.x

• add SpacyTokenizer to trained spacy_sklearn models metadata.

0.7.x to 0.8.x

• The training and loading capability for the spacy entity extraction was dropped in favor of the new CRF extractor.

Have questions or feedback?

We have a very active support community on Rasa Community Forum that is happy to help you with your questions. If you have any feedback for us or a specific suggestion for improving the docs, feel free to share it by creating an issue on Rasa NLU GitHub repository.