Skip to main content

Prerequisites

Before using AIchor, a team needs to ensure a minimum requirement for its project setup. The Machine Learning team needs to manage its code using a version control system. Currently, the system supports Github, Gitlab and BitBucket. Other VCS vendors will be supported in the future.

Launching experiments and managing their lifecycle will be driven by the Git workflow. For more details about how you can manage your code with Git we recommend reading this: a successful Git branching model.

Once your code is in a Git repository, you need to add two files to your project:

  • A Dockerfile that installs all the dependencies of your experiment scripts. You need to check that your experiment script can run manually in a docker container generated from this docker file before trying to launch it in the pipeline;
  • A manifest.yaml file that will indicate the resources needed by the experiment. The syntax of this manifest will be detailed later in this document. Once these requirements are met, the administrator of the organisation (ORG) will receive his account details in order to be able to add projects to the AIchor platform. Details on users management and experiments workflow can be found in the following sections.

Note The only technical skill required to use AIchor is that users are familiar with Docker and have their code dockerized.

Cloud providers prerequisites​

In order to leverage existing accounts on users cloud providers accounts, it is mandatory to have:

For GCP

  • GCP project ID used as input when managing engines
  • Service account with permissions allowing AIchor to import or create engines

For AWS

  • AWS Account where different resources used or created by AIchor will be created
  • Role ARN with permissions on different services used by AIchor for creation or import of engines

For Azure

  • Azure subscription where different resources used or created by AIchor will be created
  • Azure service principal Equivalent of service account used to access subscription
  • Azure Resource Group Name the project/folder within the subscription where resources are created