Elyra, a set of extensions for JupyterLab Notebooks, has reached its first major release – just three months after being presented to the general public. The project was started by IBM and is meant to give users more support when using notebooks for AI pipelines.
Since the introduction of the project, the Elyra team said it has been busy listening to what those early adopters had to say and this informed some of the useability changes in the 1.0 version. The project’s Notebook Pipeline editor, which offers a graphical approach to workflow building, now notifies users of missing or invalid configuration values and has become better at providing documentation inline.
The release is said to also handle dependencies better and helps access previously submitted experiments. In addition, Elyra can now run notebooks as batch jobs and execute Python scripts in the editor. The extensions also help to version Notebooks through an integration with version control system Git, and provide a table of contents for better navigation.
Developing code often means having to write the same lines over and over again, which is why version 1.0 comes fitted with a code snippet extension. This allows practitioners to create, edit, and store often used passages, which they can then access directly from the JupyterLab workspace for insertion into notebooks or other scripts.
In the spirit of avoiding duplication where possible, the team has also continued to improve the shared configuration service, letting various components use the same configs, and added schema-based validation capabilities as well as a full set of REST APIs to it. A new editor for browsing and changing those configurations can be found through the user interface.
To help users get up and running quicker, developers interested in Elyra can try the Binder version of the tool now. Binder is a web service that lets users open notebooks in an executable environment, by turning repositories containing Jupyter notebooks into Docker images which are then hosted at JupyterHub for easy sharing. However, there’s now also an official Elyra Docker image, which is said to make local trials easier as well.
Given that Elyra aims to “help users through the model development life cycle complexities”, as the 1.0 announcement post has claimed, the current iteration still seems to have a way to go. However, Luciano Resende, a Open Source AI Platform Architect at IBM, told DevClass in an email that “more components such as scripts and a deploy model” are planned to make their way into the project “in the near future”.