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 describes Micro-AIS, a simplified artificial immune system optimization algorithm in which antibodies represent candidate solutions. It initializes a population across the search space, evaluates each candidate with a fitness function, creates…
The article explains Deep Q-Networks (DQN) as a reinforcement learning method that uses neural networks to estimate action values across market states. It outlines the agent, environment, state, action, and reward concepts, then describes a trading setup…
This installment in a series on building an MQL5 reinforcement-learning agent reorganizes HTTP request functions into classes. The starting code uses separate procedures for GET and POST requests and a dispatcher that selects a request method. The article…
The article walks through a demonstration pipeline for forecasting financial prices: obtain EURUSD hourly data, scale closing prices, form rolling sequences, train a one-dimensional convolutional and LSTM network, export it to ONNX, and load the model in an…
The article outlines a Windows-based workflow for using R with MetaTrader 5. It connects RStudio to the terminal through reticulate and the MetaTrader 5 Python library, then uses an MT5R package and a terminal-side Expert Advisor to handle communication and…
The document describes a manual workflow for creating labeled time-series datasets from market charts. Instead of splitting a price series into separate files by trend direction, it adds a trend-group field to each observation so that the original…
This article presents a two-stage machine learning workflow for filtering Bollinger Band mean-reversion signals. A primary strategy supplies direction when price reaches an outer band; a secondary classifier estimates whether to take that trade. Its…
The article describes a machine learning workflow for searching for time-based patterns in forex trading. A filter selects bars by hour or weekday, and the labeling process keeps only examples that meet that condition. The same filter restricts trade entries…
This article explains slice sampling, an MCMC method that can draw from a target distribution using its unnormalized density. For a one-dimensional update, it samples a height beneath the density at the current point, finds an interval around that point by…
This article introduces Gaussian process classification for binary labels and explains why it requires approximate inference. A Gaussian process places a prior over latent functions, which are mapped through a sigmoid to class probabilities. Unlike…
The article describes implementing k-means clustering in MQL5, using OpenCL to parallelize computations. It motivates a native implementation for cases where Python integration limits access to terminal data, indicators, or event handling, or when clustering…
The article outlines an early data-preparation stage for a machine-learning trading project using EURUSD hourly data retrieved through MetaTrader 5. It describes constructing price and volume features such as moving averages, returns, rolling dispersion,…
The article proposes a stress test for population-based optimization algorithms: initialize every agent at the test function’s global minimum, then ask the population to find the global maximum. This removes the diversity normally supplied by random starting…
The article describes a hybrid neural network for classifying the direction of a stock’s daily candle. A one-dimensional convolutional network processes sequences of open, high, and low prices to extract features; a recurrent layer then models their temporal…
The article describes a pipeline for combining MetaTrader economic-calendar data with price bars to train and deploy a LightGBM model for news-aware forex trading. It collects event names, sectors, importance, actual, forecast, and previous values, filters…
The article introduces the Kalman filter as a way to smooth noisy price observations and estimate an underlying market movement. It explains the filter’s two stages: predicting a state from a process model, then updating that estimate using the latest…
The article explains how generative adversarial networks can augment limited financial histories with synthetic price series, including scenarios such as volatility spikes or market declines. It outlines cleaning and scaling historical exchange-rate data,…
This article outlines a Python workflow for exploring whether economic indicators can help forecast currency movements. It describes retrieving indicators such as GDP growth, inflation, interest rates, trade measures, debt, and employment from the World…
This article presents an Expert Advisor that uses readings of the Acceleration/Deceleration oscillator at several lookback offsets as inputs to a single-layer weighted sum. The sign of that score determines whether the system opens a long or short position.…
The article introduces Brent and West Texas Intermediate (WTI), describing differences in crude quality, production geography, and access to transport. It then presents a classic spread-trading premise: deviations of the Brent–WTI price spread from a…
The document explains a population-based stochastic optimizer inspired by archers aiming at a target. Each candidate solution is an agent whose coordinates are updated using another agent selected according to fitness-based probabilities. The described…
The article presents MASAAT, a portfolio optimization framework that combines multiple agents, directional price filters, and attention mechanisms. Agents use filters with different thresholds to extract trends at different scales. Cross-sectional attention…
The article builds a logistic classification model intended to identify crash conditions in stock data. It discusses candidate predictors including company earnings measures, interest rates, and inflation, while explaining that supply-demand data and current…
The article introduces the Crow Search Algorithm, a population-based method for searching an optimization space. Each candidate solution keeps a memory of its best position. During an iteration, an agent follows another agent’s remembered position by a…