A new machine learning platform that helps you quickly build industrial-grade recommendation systems

I have seen on Lakesoul's GitHub homepage DmetaSoul that they released a new open-source product, MetaSpore, related to machine learning. The function developed can help users quickly build a complete recommendation system.

MetaSpore is an open-source one-stop machine learning development platform produced by DMetaSoul, providing the whole process framework and development interface from data preprocessing, model training, offline experiment, and online prediction to online experiment bucket ABTest. It is hoped that users can quickly build industrial-grade AI systems with distributed machine learning training, high-performance model reasoning, high availability AB experimental framework, and other capabilities in a low-code way based on MetaSpore.

MetaSpore has the following characteristics:

• One-stop, end-to-end development, from offline model training to online prediction, the whole link unified development experience;
• Deep learning training framework compatible with PyTorch ecology, supporting distributed large-scale sparse feature learning
• Training framework is integrated with PySpark to read training data on data lake and data lakehouse seamlessly;
• High-performance online prediction service, supporting neural network, decision tree, Spark ML, SKLearn, and other models; Support heterogeneous computing inference acceleration;
• In offline unified feature extraction framework, automatic generation of online feature reading logic, unified feature extraction logic;
• Online algorithm application framework, providing model prediction, experimental bucket cutting flow, dynamic thermal loading of parameters, and rich debugging functions;
• Rich industry algorithm examples and end-to-end complete link solutions.

The following article will introduce several open-source products similar to Metaspore. I will also reprint the tutorials published by Metaspore, helping users build a perfect recommendation system. In addition, due to the unified software interface, The relevant algorithms, both online and offline, can be applied to real business with minimal code and configuration modifications.