Implementation of basic machine learning methods within ClickHouse using aggregate functions. This includes stochastic linear regression and logistic regression with several strategies for updating model weights.
Why it matters
This feature provides native support for commonly used machine learning algorithms in ClickHouse, enabling users to perform model training directly inside the database. It addresses the need for efficient, scalable in-database machine learning by allowing different optimization methods (simple gradient descent, momentum, Nesterov) and supports mini-batch training of customizable sizes.How to use it
Users can apply the machine learning methods via ClickHouse aggregate functions designed for stochastic linear and logistic regression. They can select the desired weight update strategy (such as simple gradient descent, momentum, or Nesterov) and specify mini-batch sizes to control training granularity within their SQL queries.