Customizing environment definitions (Watson Studio)
All environment definitions have a standard software configuration of pre-selected libraries and packages that are available when the environment runtime is started. If certain libraries or packages are missing, there are three methods that you can use to add those libaries:
- By using conda and pip directly in a notebook
- By creating a software customization in the Juypter notebook environment definition that you created
- By building a customized image
You can customize the software configuration of Jupyter notebook environment definitions that you use in the following tools:
- Notebook editor
- JuypterLab IDE
You can’t customize the software configuration of Spark, RStudio, and Hadoop environment definitions that you have created.
The following table shows the installation methods that are available for customizing the software configuration of Jupyter notebook environment definitions for specific tasks.
| Installation method | Customization type | Description |
|---|---|---|
| conda or pip in notebook | Custom libraries and Python files added through notebook | - Libraries can only be used in the notebook - Use conda rather than pip where possible for better dependency management - Packages installed from anaconda.org or IBM repositories by default - Cloud Pak for Data administrator can specify internal conda channels or a corporate proxy |
| conda or pip in environment definition | Declarative description of dependencies usable across notebooks within a project; the libraries and files are not persisted but installed when runtime is started | - Environment definition is accessible by all project members - Packages installed from anaconda.org or IBM repositories by default - Cloud Pak for Data administrator can specify internal conda channels or a corporate proxy |
| conda, pip, microdnf | New customized image is built by a Cloud Pak for Data administrator | - New custom image is built and uploaded from existing runtime image using Dockerfile - In addition to pip and conda packages, operating system dependencies can be installed using microdnf - Cloud Pak for Data administrator is responsible for maintaining image updates, including security patches |
The diagram illustrates possible software customization options for Jupyter notebook environment definitions. Custom libraries can be added through conda, pip and microdnf. The diagram shows options for accessing libraries in the public network as well as options for customizing without public access.
Next step
The default settings for conda and pip in an environment definition require that the environment runtimes have access to the public network at the time they are started. If access to the public network is not available or desired, you can customize the conda and pip configuration to access libraries by alternate methods.
Create a software customization:
- For environment definitions with public network access by:
- For environment definitions without public access by:
- For environment definitions using custom image by: