ML Model Hosting Overview

EnOS ML Model Hosting can help you quickly deploy predictive machine learning algorithm models. After an algorithm model is deployed successfully, the “Prediction” service will be created automatically and be hosted by EnOS API Management service.

The major workflow for hosting an algorithm model is shown below:

../_images/model_deployment.png

Main Functions

Model source file version control

The algorithm model source files are saved and managed through GitLab projects. When creating a deployment for an algorithm model, you can add model source files from GitLab. When updating the deployment configuration, you can add updated model files repeatedly from GitLab, thus achieving file version control.

Support for multiple running environments

When deploying an algorithm model, you can choose a running environment for the deployment, including Python 2.7, Python 3.6, and R 3.6.

Automatic creation of service

After an algorithm model is deployed successfully, the service will be created automatically and be hosted through EnOS API Management.

Algorithm model debugging

After an algorithm model is successfully deployed, you can input the test parameters and invoke the deployed service to debug the deployed algorithm model.

Deployment log information

If an error occurs when deploying an algorithm model, you can use the Error log information to troubleshoot the problem. Then, you can restart the deployment after updating model files, the main function file, or the running environment.

Restrictions and Dependencies

Data Source Registration

Before adding a model deployment file from GitLab, you need to register GitLab project data source through Data Source Registration. For more information, see Registering a GitLab Data Source.

API Management

The service automatically created after the algorithm model is deployed will be hosted by EnOS API Management. When invoking a deployed service through an application, you need to provide the service ID, business parameters, application service account, etc. For more information about EnOS APIs, go to EnOS Management Console > EnOS API.