Alink is a machine learning framework designed for real-time data processing. It uses component aggregation and resource optimization to achieve efficient data processing and resource utilization. Its innovative design enables batch task execution, minimizes intermediate data storage, and delivers superior performance in real-time computing scenarios.
Introduction
Alink is a machine learning algorithm framework and component library developed by the Platform for AI (PAI) team based on Realtime Compute for Apache Flink. Alink components can be used individually or aggregated into groups. When you configure resources for a group, its nodes can run as a batch. This process avoids persisting intermediate data to disk and improves execution efficiency and resource utilization.
Machine Learning Designer provides various Alink components, each identified by a purple mark.
Alink components available in the Machine Learning Designer component pane include Etrec collaborative filtering, FM training, FM prediction, Swing training, Swing recommendation, GraphSAGE, DeepFM, and anomaly detection components such as IForest anomaly detection, Local Outlier Factor anomaly detection, and One-Class SVM anomaly detection.
Alink component usage
In addition to their basic usage, Alink components can be grouped. When the system detects Alink nodes on the canvas that can be run as a group, you can select multiple Alink nodes by clicking the 
icon above the canvas or by using Shift+left-click. Then, right-click a blank area and select group selected nodes to perform Alink component aggregation. You can then click the 
icon to configure resources for the Alink group.
In the Custom Experiment Configuration dialog box, you can configure the Alink group name, Alink group execution resource type, Number of workers (default: 2), and Memory per worker (MB) (default: 4096).
An Alink group's configuration overrides the configurations of its individual components.