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 article describes a neural-network trading workflow using Python, TensorFlow, and Keras, with MQL5 scripts supplying historical prices and indicator data. The system separates buy and sell examples and uses two networks for each direction: one learns to…
The article explains SAC+DICE, an off-policy reinforcement learning approach that combines optimistic exploration with correction for the mismatch between replay-buffer samples and the current policy. It motivates the method by noting that conservative…
The article concludes a series on GinAR, a neural architecture for noisy financial time series. Its cells combine contextual attention, adaptive graph processing, recurrent state, and separate forget and reset controls. Multiple layers process a sequence,…
This article sketches an experimental approach to market analysis using Qiskit and MetaTrader 5. It describes encoding price movements into quantum circuits, applying quantum phase estimation, and combining quantum methods with classical probability and…
The document introduces DUET, a neural forecasting framework for multivariate time series that clusters information along both time and variable channels. Its Temporal Clustering Module groups temporal patterns and uses a mixture of encoders with noisy…
This tutorial describes moving a small financial language model training example from CPU to GPU, with emphasis on preparing AMD hardware for ROCm. It recommends Ubuntu because the article reports that ROCm-enabled PyTorch was not available for Windows or…
This article explains bidirectional piecewise linear representation (BPLR), a way to compress a time series into line segments while retaining its broad shape. For anomaly detection, the described method divides cyclic data into non-overlapping subsequences,…
This article introduces support vector machines as supervised classifiers that map labeled input examples into a feature space and determine a boundary separating two categories. It uses a two-dimensional diagram explanation and a hypothetical…
This tutorial walks through setting up Python for use with MetaTrader 5, implementing a perceptron and a multilayer perceptron, and connecting Python workflows with MQL5. It explains network initialization, forward propagation, sigmoid activation,…
The article applies category-theory naturality squares to classify price-change data across three linked currency pairs. USDJPY forms the source series, while EURJPY and EURUSD form the paired destination series. The proposed construction connects their…
This article explains the Encoder design for a probabilistic time-series model that combines latent linear dynamics, error analysis, and adaptive uncertainty estimation. Its KoopmanNet component predicts latent-state transitions and reconstructs earlier…
This article introduces K²VAE, a framework intended to forecast distributions of future time-series paths rather than a single price estimate. It combines a variational autoencoder with a learned Koopman representation, which approximates nonlinear dynamics…
This article explains how ADX and CCI can be paired as inputs to a supervised multilayer perceptron. ADX measures trend strength without indicating direction, while CCI tracks price momentum relative to a typical-price average. It outlines Python…
This article proposes treating consecutive price changes as angles and using sequences of those angles as features for market prediction. It frames the idea as a more systematic alternative to manually drawing Gann angles, with the goal of capturing…
This article builds a native MQL5 state store for Expert Advisors that need to resume after a restart. To avoid corrupting the active file during an interrupted save, it writes new data to a temporary file, flushes and closes it, then moves it into place. A…
The document describes the Cricket Algorithm, a population-based optimization method inspired by cricket calls and assembled from mechanisms associated with the Bat, Particle Swarm, and Firefly algorithms. Candidate solutions update velocity and position…
The article presents the Artificial Atom Algorithm (A3), a population-based metaheuristic for optimizing strategy parameters. Candidate solutions are represented as atoms, with each coordinate corresponding to a decision variable. The described procedure…
The article explains PointNet, a neural-network architecture designed to process unordered point sets directly. Rather than converting a point cloud into a voxel grid or image, it applies shared point-level transformations and uses max pooling to create a…
The article explains Differential Search Algorithm (DSA), a population-based optimizer inspired by migration. Candidate solutions move through a search space using one of several direction rules: toward a random peer, a selected top solution, a shared…
This installment describes connecting a trained classifier to MetaTrader 5 for live order execution. Predictions above or below a threshold trigger long or short trades with preset stop-loss and take-profit levels, while a cap on open positions limits new…
The article explains how to extend a simple linear regression implementation to handle multiple independent variables using matrix algebra. It introduces the design matrix, including a column of ones for the intercept, and describes transposing data,…
The article outlines a machine learning workflow that combines volume analysis with price features to predict market movements. It proposes rolling volume averages and ratios, price and volume momentum, volume volatility, and price-volume correlation.…
The article compares three ways to combine population-based optimizers: merge their search logic, run them in successive stages, or run them in parallel and combine their best candidates. Its worked example pairs Grey Wolf Optimization with the Cuckoo…
The article describes an end-to-end workflow that connects MetaTrader 5 data collection to Python model training and live inference. Historical bars are uploaded in chunks, enriched with technical and statistical features, labeled, and stored for training. A…