Gaussian Process Kernels and Online Trading Indicators in MQL5

Summary

This article completes an MQL5 Gaussian process library by describing interfaces for covariance kernels, likelihoods, and inference. It details RBF, linear, and periodic kernels, including their hyperparameters and covariance derivatives, then applies the library to synthetic data and online classification and regression indicators that retrain on each new bar. The classifier estimates one-step-ahead price direction; the regressor forecasts price with a confidence interval.

The article presents implementation details and identifies performance and scaling limits. It reports that optimization was relatively slow with the MinBleic optimizer and points to faster gradient optimization, improved Newton-mode search for Laplace approximation, and sparse methods as possible extensions. The excerpt does not provide quantitative trading results or enough detail to assess out-of-sample performance. Its indicators demonstrate how to apply Gaussian processes in a trading platform, but retraining each bar and the stated computational limitations may constrain practical use.

Key ideas

• Kernel interfaces make Gaussian process components extensible across covariance, likelihood, and inference implementations.
• The RBF kernel models smooth functions, the linear kernel represents linear relationships, and the periodic kernel captures recurring patterns.
• Kernel hyperparameters can be optimized using derivatives of the covariance matrix.
• The examples use online classification and regression indicators that retrain on each new bar.
• The article identifies optimizer speed, Laplace approximation convergence, and dense matrix scaling as areas for improvement.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.