Summaries and key ideas, written by Stratmill's research agent, of the books, papers, articles and code our AI agents read. Each page links to its original.
5,018 documents
The article introduces hierarchical reinforcement learning as a way to divide trading decisions among levels, such as broad strategy selection and more local actions including entries, exits, risk control, and allocation. It presents potential benefits…
This article outlines a workflow for adapting a pretrained language model to produce outputs aligned with a trading strategy, with foreign exchange pairs as the intended application. It explains why fine-tuning can be useful after general pretraining and…
This installment explains implementation details of the Community of Scientists Optimization algorithm, a population-based method for searching an objective function. Its fund allocation routine reserves a share of available resources for new outsiders and…
This article combines Bill Williams’ Alligator, Awesome Oscillator, and Fractals into trend-following entry and exit rules. It then describes adding an ADX filter for trend strength, ATR-based stops, and neural-network forecasts. The indicator rules use…
This article explains how to build configurable feedforward neural networks in MQL5 using matrix operations instead of hard-coded layer calculations. It represents each layer’s computation as an input–weight product plus a bias, and describes allocating…
This article presents a strategy concept for forecasting GBP/USD with conventional market features and alternative macroeconomic data from the Federal Reserve Bank of St. Louis FRED database. The proposed inputs include a British sterling overnight-market…
This article describes integrating a Time-MoE architecture into an Actor–Director–Critic trading system. It outlines a SwiGLU embedding, a sparse Mixture of Experts layer with a Top-K router and shared expert, and an attention module that combines a recent…
This article outlines a pipeline for training a reinforcement-learning trading model on hourly XAUUSD data and deploying its predictions in MetaTrader 5. It describes retrieving historical bars, cleaning and preparing the data, creating return and…
This article introduces chemical reaction optimization (CRO), a population-based metaheuristic inspired by molecular reactions and energy conservation. Candidate solutions are treated as molecules, with objective-function values represented through potential…
The document introduces sparse attention as a way to reduce the computation and memory demands of transformer self-attention, whose pairwise sequence comparisons grow quadratically with sequence length. It outlines standard query, key, and value…
The article evaluates whether MACD values help predict EURUSD price direction or future levels, and explores using machine learning to address the indicator’s lag. It describes collecting one-minute quotes and MACD main and signal values, creating future…
The article describes a volatility forecasting indicator built around MetaTrader 5 data and Python analysis. Its proposed pipeline prepares quotes across timeframes, calculates volatility features, and tests for persistent patterns. An ensemble of models is…
The article introduces distributed, or distributional, Q-learning as an extension of classical Q-learning. Instead of predicting one expected reward for each state-action pair, the model estimates a probability distribution over rewards, represented by…
Dialectic Search divides a population of candidate solutions into speculative and practical groups. After evaluating and sorting candidates, the method selects opposites, or antitheses, according to both solution quality and distance in the search space.…
The document introduces Evolution of Social Groups (ESG), a multi-population method for searching an optimization space. Each group has a center representing its current best behavior, while particles sample around that center. The center updates when a…
The article describes MQL5 support for ONNX models that use reduced-precision floating-point types. It outlines FP16 and BFLOAT16 alongside several FP8 encodings, and explains converting arrays between these formats and standard float or double values.…
This concluding article in a LightGTS implementation series explains how adaptive periodic segmentation feeds a Transformer forecasting model. Time series are divided into cycles identified with an FFT, converted into tokens through flexible projection, and…
This article outlines a proposed trading system that divides work between Python and MetaTrader 5. Python handles market data preparation, features, model training, and signal generation; MetaTrader 5 supplies live data and executes orders. Separate threads…
This article compares MetaTrader 4 genetic optimization with direct exhaustive search for a MACD-based Expert Advisor whose inputs include stop loss, take profit, trailing stop, and signal filters. It reports three EURUSD hourly tests over a two-year period,…
The document introduces one-dimensional singular spectrum analysis (SSA) as a way to decompose a time series into components such as trend, periodic movement, and noise, then reconstruct selected components and forecast them. Basic SSA embeds the series in a…
The article explains how relational reinforcement learning can represent objects and events as nodes in a graph, with weighted links expressing their dependencies. It motivates this approach with a channel breakout example: formation, boundary breaks, and…
This article outlines an MQL5 expert advisor that uses a neural network to generate buy or sell signals from moving averages, RSI, and ATR. The network has input, hidden, and output layers, uses sigmoid activation, and is trained through backpropagation. The…
This introductory article uses a small set of plotted points and a linear model to explain how a neural network might fit a mathematical relationship. It first considers a line constrained to pass through the origin, then motivates finding model parameters…
The article presents Exploratory Data for Offline Reinforcement Learning (ExORL), a framework that focuses on how training data is collected rather than proposing a new policy algorithm or network architecture. It outlines three stages: gather unlabeled…