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 presents a temporal nested cross-validation design intended to reduce optimistic evaluation in financial machine learning. It divides observations into an outer training zone, an inner validation checkpoint, and a final test segment that is…
The article outlines a workflow for using an LSTM regression model to estimate future trend strength and filter trend-following entries. It motivates forecasting ADX rather than relying on its current value, which may rise after much of a move has already…
The article describes a trading signal workflow that encodes selected Ichimoku and Wilder’s ADX conditions as binary pattern flags, then trains a β-variational autoencoder to learn compact representations. Its motivation is that scaling continuous indicator…
The article explains Dynamics-Aware Discovery of Skills (DADS), an unsupervised reinforcement learning method intended to learn reusable behaviors that are both diverse and predictable. It contrasts DADS with DIAYN: DIAYN encourages distinguishable behavior,…
The article implements a version of the DADA time-series anomaly detection framework, focusing on its adaptive bottleneck module. The design uses multiple autoencoder experts with different latent compression sizes. A gating mechanism selects the most…
The document explains FreDF, a training approach for direct multi-step time-series forecasting. Direct forecast models predict several future values at once and commonly calculate their loss step by step in the time domain. The article argues that this can…
This article outlines an MQL5 Expert Advisor architecture that periodically rebuilds and trains a multilayer perceptron, then uses its forecasts while later optimization runs. It describes a workflow of creating the network, preparing and normalizing input…
The article describes FinMem, an LLM-based trading-agent design organized around profiling, memory, and decision-making. Its profile combines company and sector context with a selected risk style: risk-seeking, risk-averse, or self-adaptive. The memory…
ConserWeightive Behavioral Cloning (CWBC) is presented as a way to make offline, return-conditioned behavior-cloning policies more reliable when training trajectories vary in quality. The method has two parts: weight trajectories to give higher-return…
ConserWeightive Behavioral Cloning (CWBC) is presented as a way to make offline, return-conditioned behavior-cloning policies more reliable when training trajectories vary in quality. The method has two parts: weight trajectories to give higher-return…
This article explains an MQL5 trade filter that uses online logistic regression to estimate whether a base signal will win. It constructs scaled features from RSI, EMA distance, relative ATR, candle shape, MACD, and volume, then scores an EMA crossover…
The article describes a trading-oriented implementation of ACEFormer, combining adaptive empirical mode decomposition with a Transformer forecasting architecture. ACEEMD is used to reduce high-frequency noise while retaining turning points; the resulting…
The article explains how four entropy measures—Shannon, Plug-In, Lempel-Ziv, and Kontoyiannis—can describe the distribution and ordering of price-direction changes within a bar. It turns bid movements into a three-symbol sequence, then computes each measure…
The article describes a proposed automated system for cryptocurrency CFDs across separate MetaTrader 5 accounts. Its architecture assigns each account an independent Python process and terminal, with shared configuration values and trade execution handled by…
This article examines why explainability methods can give conflicting accounts of the same machine learning model. It distinguishes global explanations, which describe overall feature importance or behavior, from local explanations focused on an individual…
The article proposes a market prediction pipeline combining quantum circuit measurements, CatBoost, and a language model. It motivates the design with shortcomings of conventional indicators and reports that an LLM’s analysis changed when prompts or…
This article describes the Bird Swarm Algorithm, a population-based optimization method inspired by flocking and feeding behavior. It represents candidate solutions as birds that can search as producers, follow producers as beggars, feed by combining…
This article explores how to inspect a CatBoost gradient-boosted tree model trained to forecast a future price from market data and technical indicators. It motivates feature interpretation as a way to debug a model, evaluate engineered inputs, guide future…
This tutorial introduces a single artificial neuron using a small set of input-output examples whose relationship is multiplication by two. It presents neural networks as mathematical models that estimate patterns from data, and compares training,…
This article develops a predictive modeling workflow in MQL5 using market price data and matrix operations. It normalizes rows of open, high, low, and close observations, separates training and test samples, and estimates linear-model coefficients with a…
This installment of an MQL5 tutorial extends a ChatGPT-connected trading assistant with longer multiline prompts, a sidebar for saved conversations, and initial signals informed by chart data. It describes modularizing the program, refining text display and…
This article introduces simple recurrent neural networks (RNNs) for forex time-series forecasting and explains how their hidden state carries information across sequential inputs. It outlines the recurrent computation and backpropagation through time,…
This article outlines a workflow for running a local language model with Ollama and adapting it to produce structured trading signals. It describes downloading a small base model, defining trading rules and output fields in a model configuration, and testing…
The article explains a reinforcement learning approach that trains a neural network to choose trading actions directly. Unlike Deep Q-Learning, which estimates expected rewards for actions, the policy model outputs a probability distribution over actions.…