Deploy a pipeline as an online service
Machine Learning Designer lets you package a batch data-processing pipeline—covering data pre-processing, feature engineering, and model prediction—into a pipeline model, then deploy it to Elastic Algorithm Service (EAS) as an online service.
Only pipelines built entirely with Alink algorithm components (marked with a small purple circle in the canvas) can be deployed as online services.
Limitations
| Constraint | Details |
|---|---|
| Component type | Only Alink algorithm components are supported. They are marked with a small purple circle in the canvas. |
| Run status | All paired training and prediction components in the pipeline must have completed successfully (marked with a green check) before packaging. For example, to deploy a linear regression model, both the Linear Regression Training and Linear Regression Prediction components must show a green check. |
| Topology | Online services accept only single input and output. Select a single serial link from the Directed Acyclic Graph (DAG) of your batch pipeline. |
Prerequisites
Before you begin, ensure that you have:
A batch data-processing pipeline that implements data pre-processing, feature engineering, and model prediction, and has been run successfully. For more information, see Build a model
Deploy a pipeline as an online service
Go to the Machine Learning Designer page.
Log on to the Machine Learning Platform for AI console.
In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the workspace you want to manage.
In the left-side navigation pane, choose Model Training > Visualized Modeling (Designer).
On the Pipelines tab, double-click the pipeline to open it.
In the top navigation bar of the canvas, choose Create Pipeline Model.
Select the nodes that form a serial data processing link, then click Next. A serial link typically consists of 1 to N prediction components. For example, the following link normalizes data, applies one-hot encoding, aggregates vectors, and then runs the FM Prediction component.
Selecting a node that forms a serial link with its upstream or downstream nodes automatically selects those connected nodes.
Unselecting a node automatically unselects all nodes that were selected with it.
In the Create Pipeline Model dialog box, click Next to start packaging.
The system packages the data prediction link and its models into a pipeline model. A batch task whose name starts with model-combination- is launched automatically. To monitor progress, click View All Tasks in the top navigation bar. In the Previous Tasks dialog box, find the task and check its status. Packaging takes approximately 3 to 5 minutes. Wait until the task state changes to Succeeded before proceeding.In the Previous Tasks dialog box, find the task and click Model in the Actions column.
This opens the EAS-Online Model Services page. For more information about completing the deployment, see Model service deployment by using the PAI console and Machine Learning Designer.