Build a KMeans clustering pipeline over a CSV dataset using PyAlink
Run classification or association rule mining like Apriori on Flink data
Combine several feature engineering steps into one reusable pipeline
Process the same algorithm pipeline in either batch or streaming mode
| alibaba/alink | nekogram/nekogram | litesuits/android-common | |
|---|---|---|---|
| Stars | 3,616 | 3,612 | 3,624 |
| Language | Java | Java | Java |
| Setup difficulty | hard | moderate | easy |
| Complexity | 4/5 | 3/5 | 2/5 |
| Audience | data | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires matching Python, Java 8, and Flink versions precisely to the PyAlink or Maven package you install.
Alink is a general purpose machine learning algorithm library built on top of Apache Flink, developed by the PAI team at Alibaba's computing platform division. Flink is a framework for processing large amounts of data, either as one large batch or as a continuous stream, and Alink adds a set of ready made algorithms on top of it for tasks like classification, clustering, feature engineering, and association rule mining such as the Apriori algorithm. The project can be used from Python, through a package called PyAlink, or directly from Java. Different PyAlink packages exist for different versions of Flink, from 1.9 up to 1.13, and each is installed with a matching pip command. Using it from Python requires Python 3.6, 3.7, or 3.8 and a Java 8 installation on the same machine. The Java version is added to a project through Maven, with a dependency block that also needs to match the Flink version in use. In both languages, the pattern is the same: you build a pipeline out of algorithm components, each created with a constructor and configured with setter methods, then connect components together and run them against a data source. An example in the README loads a CSV file of flower measurements, assembles several numeric columns into a single feature column, and feeds that into a KMeans clustering step to group the rows, printing the result. Most of the project's documentation, tutorials, and component list are hosted on a separate website and are written in Chinese, with an English-language version of the README also available in the repository. The README itself is sparse on English detail beyond installation and one basic usage example, so anyone wanting the fuller catalog of available algorithms will need the linked documentation site.
A machine learning algorithm library from Alibaba built on Apache Flink, offering classification, clustering, and feature engineering from Python or Java.
Mainly Java. The stack also includes Java, Apache Flink, Python.
No license terms are stated in the explanation provided.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly data.
This repo across BitVibe Labs
Verify against the repo before relying on details.