Articles on machine learning in trading

We invite you to explore the practical implementation of a sparse mixture of experts block for time series in the OpenCL computing environment. This article provides a step-by-step explanation of how masked multi-window convolution works, as well as how gradient-based training is organized in the presence of multiple information streams.

In today's article, we will see how a simple code change that makes a neuron slightly more specialized can significantly speed up the training stage. After all, once a neuron or neural network, as we will see later, has been trained, the work it performs becomes much faster. We will also discuss a problem that exists but is rarely mentioned.

We continue our acquaintance with the Mamba4Cast framework. Today, we will delve into the practical implementation of the proposed approaches. Mamba4Cast was designed not for lengthy warm-up on every new time series, but for immediate deployment. Thanks to the concept of Zero-Shot Forecasting, the model can produce high-quality forecasts on real-world data without additional training or hyperparameter tuning.

A new original population-based algorithm, ECEA, is presented. Inspired by the process of water freezing, it adapts ideas from the Crystal Energy Optimizer (CEO) algorithm, which uses graph-based search, for general optimization problems. The algorithm uses a dynamic elite group, three search strategies, and a periodic diversification mechanism.

An MQL5 port of four entropy estimators — Shannon, Plug-In, Lempel-Ziv, and Kontoyiannis — operating on the intrabar tick-rule sequence. CopyTicksRange() limits data to the broker's cached tick window, so features apply to recent bars only. The implementation encodes bid-direction ticks from MqlTick, replaces NumPy-dependent steps with array-based methods, and ships CEntropyFeatures.mqh and EntropyViewer.mq5 for EA and indicator use.

This article presents a reproducible MetaTrader 5 workflow: collect history, engineer nine context features, label simulated EMA crossover trades, train with FLAML, and export to ONNX with fixed opset and plain probabilities. The Expert Advisor loads the model natively, mirrors the Python feature contract, and uses a tunable confidence threshold as a trade filter. Readers can swap signals and features to reuse the same pipeline.

This article performs a numerical verification of MQL5 eigendecomposition for a covariance matrix using the spectral theorem A = V Λ Vᵀ. It reconstructs the matrix with Diag(), Transpose(), and MatMul(), computes the residual and its Frobenius norm, and shows that deviations remain at floating‑point precision, with results printed to the Experts journal.

The article focuses on the practical implementation of the TimeFound model for time series forecasting. The key stages of implementing the framework's main approaches using MQL5 are examined.

This article implements BOSS from scratch in MQL5 and applies it to regime classification: SFA turns windows into words, bags record word frequencies, and an ensemble over window lengths votes on labels. We cover the encoding steps, the BOSS distance, training with auto-generated regime labels, and practical parameters. A BTCUSD benchmark versus DTW shows higher macro accuracy on clean data and markedly faster inference.

This article builds a white-box classifier in MQL5 using the Tsetlin Machine. It learns human-readable AND-rules instead of weights, trains with integer state updates, and requires no external dependencies. You will assemble the automaton, clause, and multi-class voter, verify on XOR and other boolean tasks, booleanize indicators, label by forward ATR-scaled return, save the model to CSV, and view active rules on a live chart.

The article provides a fascinating look at how SwiGLU embedding reveals hidden market patterns, and how a sparse Mixture of Experts within a Decoder-Only Transformer makes forecasts more accurate at reasonable computational cost. We take an in-depth look at the integration of Time‑MoE into MQL5 and OpenCL, and provide a step-by-step guide to configuring and training the model.

This article performs a numerical verification of MQL5 eigendecomposition for a covariance matrix using the spectral theorem A = V Λ Vᵀ. It reconstructs the matrix with Diag(), Transpose(), and MatMul(), computes the residual and its Frobenius norm, and shows that deviations remain at floating‑point precision, with results printed to the Experts journal.