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
This educational article explains how to fit a straight-line equation to paired data by estimating both its slope and intercept. It derives the least-squares solution from sums over observed points, then shows how the resulting calculations can be…
This article implements a version of the FinMem trading-agent architecture in MQL5. Its memory design separates short-term market information from longer-lived patterns or strategies. The described neural module processes market-state tensors and account…
The article compares adaptive learning rate methods for multilayer perceptrons with a one-cycle schedule. It describes adaptive gradient, RMS, mean exponential, and delta approaches, which derive parameter-specific updates from training gradients while…
The article uses a Darvas box breakout as a case study for supervised machine learning in algorithmic trading. It distinguishes generating trade signals with a model from using a model merely to filter trades produced by an existing strategy. The workflow…
The article describes an MQL5 money management class that combines a one-level Haar wavelet filter with an LSTM network to scale position size in volatile, trend-oriented markets. The wavelet step transforms recent log returns into approximation and detail…
This tutorial introduces gradient descent as an iterative way to reduce a model’s cost function. It explains differentiating a function to obtain its gradient, then updating a parameter in the direction opposite that gradient. The learning rate sets the step…
This article presents practical components of an Elite Crystal Evolution Algorithm, a population based optimizer inspired by crystal formation. The population is divided into elite and regular agents. Elite agents conduct local search, while regular agents…
The article proposes combining balance-sheet data from the Federal Reserve, European Central Bank, Bank of Japan, and People’s Bank of China into a global liquidity measure for forecasting currency movements. It explains the intuition that balance-sheet…
The document explains the Black Hole Algorithm (BHA), a population-based method for optimizing continuous or discrete search problems. It initializes candidate solutions as randomly placed stars, selects the current best candidate as the black hole, and…
The article examines how learning-rate choices affect a simple generative adversarial network used as an MQL5 trading signal. It describes fixed and scheduled rates, including step decay, exponential decay, and cyclical adjustment. In broad terms, a fixed…
This article describes Goal-Conditioned Predictive Coding (GCPC), a two-stage approach to offline reinforcement learning. It separates trajectory representation learning from behavior policy learning: a trajectory model is pretrained to reconstruct masked…
The article describes an enhanced stepwise feature-selection method for supervised learning. Instead of retaining only one current feature set, it keeps several high-performing subsets and tests candidate additions against them, aiming to capture useful…
The document explains two population-based optimization methods: (μ,λ)-ES, where offspring replace their parents, and (μ+λ)-ES, where parents and offspring compete for the next generation. Both use mutation and selection to search a real-valued parameter…
This programming tutorial compares queues, linked lists, and trees, emphasizing how their structures affect insertion, removal, and search. Queues suit ordered processing but make arbitrary middle changes awkward. Linked lists allow efficient insertion and…
This article builds a two-model forecasting approach for NZDJPY: an initial model predicts a future close, and a second model estimates the initial model’s residuals to correct its forecast. The author describes collecting minute-level prices, exploring…
The document introduces FinCon, a proposed system that models an investment team with specialist analyst agents reporting to a manager. Analysts process different kinds of market information; the manager combines their findings, makes trading decisions, and…
This article extends a brute-force method for finding trading patterns by dividing historical quotes into time-based samples. The proposed segments include days of the week and intraday windows, with further discussion of months and years. Segmentation is…
The article reviews Generalized 3D Referring Expression Segmentation, a model that uses multiple learned queries to identify zero, one, or several matching objects in a point cloud from a text description. Its components include text-guided sparse query…
The article shows how to organize multiple ONNX classifiers in MQL5 using a shared base class and model-specific subclasses. The base class stores the trained symbol and timeframe, checks that an expert advisor is using the matching market, creates and…
This article explains Stochastic Diffusion Search (SDS), a population optimization method in which agents hold candidate hypotheses, perform inexpensive partial evaluations, and share promising hypotheses directly with other agents. Restaurant-choice and…
This article covers an implementation step for SAMformer, a shallow Transformer approach to multivariate time-series forecasting that uses Sharpness-Aware Minimization (SAM). It focuses on adding SAM optimization to a convolutional layer in an MQL5…
The article introduces Superpoint Transformer (SPFormer), an end-to-end method for segmenting individual objects in sparse 3D point clouds. A sparse 3D U-Net first extracts point features, which are pooled into superpoints to reduce the representation size.…
This article describes a custom MQL5 Wizard signal class that pairs four B-Tree-inspired search modes with an optional Bayesian neural network. The proposed modes represent direct lookup, a local range average, a search for the largest recent price change,…
The article explains SEFormer, a Transformer design for 3D point-cloud detection that adds spatial structure to attention. Standard convolution encodes direction and distance through its kernels but aggregates neighbors uniformly; ordinary self-attention can…