Environments¶
Deployment versions and training experiments make use of environments. Environments are build up of a base environment and files specifying any additional dependencies. Whenever you create a new deployment version with dependency information in the deployment package, a custom environment is automatically created for you based on the provided information and whatever base environment you selected.
When you create a new training experiment, you can specify what environment should be used for your experiment, or you can create a new custom environment. When creating a new custom environment you need to specify what base environment to use, and upload files detailing what dependencies should be installed in your custom environment. How to specify the latter is dependent on what programming language you use:
In the case of Python you can specify your dependencies with a zipfile containing:
- a
requirements.txt
for specifying what pip packages to install - a
ubiops.yaml
for specifying what OS level packages to install
In the case of R you can specify your dependencies with a zipfile containing::
- an
install_packages.R
for specifying what R packages to install or anrenv.lock
for specifying what R environment to restore withrenv::restore()
- a
ubiops.yaml
for specifying what OS level packages to install
After uploading the necessary files and clicking create, your environment will start building. Please make sure you pass your dependency files as a zip, rather than individual files. Uploading an unzipped dependency will result in an error. If there is anything wrong with the uploaded files the build of your environment will fail and it will not be possible to use it. Any errors that occur during building can be found in the logs.