Designer integrates seamlessly with Elastic Algorithm Service (EAS). After you complete your offline training, prediction, and evaluation pipelines, you can deploy a single model to EAS to create an online service. This topic describes how to deploy a model trained in Designer to EAS by using one-click deployment or manual deployment.
Prerequisites
You have trained a model and verified its accuracy. For more information, see Build and debug a model.
One-click deployment for a single model
Supported model components
The following model components support one-click deployment to EAS. Some other model components support only manual deployment. For more information, see Manual deployment for a single model.
|
Component name |
Format |
EAS processor |
Description |
|
Logistic Regression for Binary Classification |
PMML |
PMML |
Before training, click the model component and select Generate PMML on the Field Settings tab. |
|
GBDT Binary Classification |
PMML |
PMML |
|
|
Linear SVM |
PMML |
PMML |
|
|
Logistic Regression for Multiclass Classification |
PMML |
PMML |
|
|
Random Forest |
PMML |
PMML |
|
|
Naive Bayes |
PMML |
PMML |
|
|
K-means Clustering |
PMML |
PMML |
|
|
GBDT Regression |
PMML |
PMML |
|
|
Linear Regression |
PMML |
PMML |
|
|
Scorecard Train |
PMML |
PMML |
|
|
Text Summarization |
tgz package |
EasyNLP |
The PAI-provided EasyNLP processor is hosted in a public Object Storage Service (OSS) bucket and is configured automatically, requiring no custom setup. |
|
Image Classification (torch) |
tgz package |
EasyCV |
The PAI-provided EasyCV processor is hosted in a public Object Storage Service (OSS) bucket and is configured automatically, requiring no custom setup. |
|
PyAlink Script |
AlinkModel |
Alink |
For more information, see PyAlink Script. |
|
XGBoost Train |
XGBoost |
XGBoost |
For more information, see XGBoost Train. |
Procedure
-
Go to Visualized Modeling, select a workspace, navigate to the Designer page, and then open your target pipeline.
-
Above the pipeline canvas, click Model List. The system automatically detects deployable models generated by the pipeline.
In the Model List dialog box, select a target model, such as Logistic Regression for Binary Classification-1. You can click Register Model to add the model to AI assets for versioning, or click Deploy to EAS to create a model service.
-
Select the target model and click Deploy to EAS to open the service creation page.
-
Configure the parameters and deploy the model service.
The system automatically matches the Model File and Processor Type. For more information about other parameters, see Custom deployment.
-
Click Deploy. When the Service Status changes to Running, the deployment is complete.
Manual deployment for a single model
The following model components do not support one-click deployment. After model training is complete, you must use the Export a General-purpose Model component to assemble the model, export it to an OSS directory, and then deploy it manually.
|
Component name |
Format |
EAS processor |
Procedure |
|
PS-SMART Binary Classification Training |
PS format |
PS algorithm |
Connect an Export a General-purpose Model component downstream from this component. |
|
PS-SMART Multiclass Classification Training |
|||
|
PS-SMART Regression Training |
After you export the model to an OSS bucket, you can manually deploy the model to EAS. For more information, see Custom deployment.
FAQ
When I use one-click deployment, a supported node is dimmed and cannot be selected. In the Model List dialog box, only the K-means Clustering-1 model is displayed, and the only available button is Close, without any options to register or deploy. How can I fix this?
Click the component node, select Generate PMML on the Field Settings tab on the right, and then rerun the node.
Related documents
-
You can go to the PAI-EAS online service page to view service status or perform management operations. For more information, see Service deployment.
-
You can use the online debugging feature to test whether the service runs as expected. For more information, see Online debugging of a service.
-
After a model service is deployed, you can use the Update EAS Service (Beta) component in Designer to update the service on a schedule. For more information, see Update online model services on a schedule.