Online Learning

OnlineLearning.jl provides an implementation of online mini-batch learning algorithms for prediction tasks in Julia. It enables incremental model updates on streaming or large datasets through mini-batch optimization, making it suitable for scenarios where data arrives sequentially or doesn't fit in memory.

Key Features

• GLM Learners — Generalized linear models for regression and classification, including linear, logistic, and quantile regression variants
• Regularization Support — L₁ and L₂ regularization through GLMNetLearner for feature selection and preventing overfitting
• Multiple Optimizers — Various stochastic gradient descent implementations including SimpleSGD, AdaDelta, AdaGrad, and AveragedSGD
• Sparse/Dense Data — Support for both dense and sparse feature matrices through DSMat abstraction
• Incremental Updates — Models can be updated incrementally with mini-batches via the update! method
• SVM Implementation — Support vector machine learner (partially implemented)

Philosophy

The project focuses on providing efficient, incremental learning algorithms that can handle both dense and sparse data structures while maintaining flexibility through composable learner and optimizer components.