Article-22220-Real-Time-Entropy-Trading-System

This repository is an article-derived reference project based on the original MQL5 article. It does not claim to reproduce the full original source code unless files are explicitly attached.

Overview

This project documents a hybrid MQL5 + Python trading system centered on real-time market entropy analysis. The architecture combines:

• an MQL5 Expert Advisor for tick collection, HTTP communication, and trade execution
• a Python feature-engineering and model-training pipeline
• a Flask inference server for live signal generation

The article describes how Shannon entropy, volatility metrics, RSI, and trend features are transformed into an 8-feature input vector for a neural network that outputs directional probability. A volatility regime detector then adjusts confidence thresholds and trading risk parameters.

Original Article

• Article ID: 22220
• Author: Hlomohang John Borotho
• Publication date: 2026.05.19
• Category: Experts
• URL: https://www.mql5.com/en/articles/22220

Repository Purpose

This repository should be treated as a reference or reconstruction project derived from the article content. Its purpose is to preserve the article’s technical structure and described file layout for:

• studying entropy-based tick-level trading workflows
• understanding MQL5-to-Python inference integration via HTTP
• reconstructing the Python training/inference stack and the MQL5 EA
• serving as a base for further experimentation with adaptive risk logic

Key Concepts

• Shannon entropy on rolling return windows
• volatility entropy from squared returns
• adaptive volatility regime detection
• neural-network directional classification
• feature scaling with StandardScaler
• Flask-based local inference API
• MQL5 WebRequest communication
• adaptive lot sizing, SL, and TP management
• tick-level execution rather than candle-close logic

Algorithm / Architecture Summary

1. Historical data collection

• Python connects to MetaTrader 5 and downloads historical XAUUSD H1 data.
• Data is saved to CSV for offline training.

2. Feature engineering

• Compute log returns from recent prices.
• Derive normalized entropy and volatility entropy using percentile-based bins.
• Compute volatility statistics such as standard deviation, MAD, range, skewness, and kurtosis.
• Estimate trend slope and regression R².
• Build an 8-element feature vector including entropy, return statistics, trend metrics, skewness, and normalized RSI.

3. Regime detection

• Maintain entropy history in rolling buffers.
• Classify market state as LOW_VOLATILITY, NORMAL, HIGH_VOLATILITY, or EXTREME_VOLATILITY.
• Produce risk multipliers for stop-loss, take-profit, and position sizing.

4. Model training

• Use a PyTorch feed-forward network with batch normalization, ReLU, dropout, and sigmoid output.
• Train on rolling-window samples labeled by future return direction over a fixed horizon.
• Save the fitted scaler and best model weights.

5. Live inference

• MQL5 EA collects recent tick bid prices and computes RSI.
• EA sends JSON to a local Flask /predict endpoint.
• Flask rebuilds features, detects regime, runs model inference, and returns:
• probability
• entropy metrics
• signal
• regime
• volatility multiplier
• confidence

6. Execution

• EA validates confidence, cooldown, current exposure, and reversal rules.
• Orders are placed through CTrade.
• Open positions may have SL/TP adjusted when the regime changes.

Mentioned or Attached Files

Explicitly attached files

The article states that the following files are contained in the archive:

• GettingHistData.py — downloads historical XAUUSD H1 data from MetaTrader 5
• Features.py — feature engineering and volatility regime detection
• Model.py — PyTorch neural network definition
• Train.py — training pipeline and scaler/model persistence
• Server.py — Flask inference server
• Entropy.ipynb — notebook for step-by-step execution
• Real-Time Entropy.mq5 — MQL5 Expert Advisor

Files mentioned in the text

• XAUUSD_H1.csv — generated dataset used by the training script
• entropy_model.pth — saved PyTorch model weights
• scaler.pkl — saved feature scaler

Statistics

• Article sections described: 6 main sections
• Python components described: 5
• MQL5 components described: 1 EA
• Model input features: 8
• Primary rolling window size: 50
• Training horizon: 5
• Flask endpoint: /predict
• Health endpoint: /health

Tags

MQL5, MetaTrader5, Python, Flask, PyTorch, Machine-Learning, Entropy, Volatility-Regime, Algorithmic-Trading, Expert-Advisor

Difficulty

Advanced

Requires familiarity with:

• MQL5 EA development
• Python data processing
• PyTorch model training
• REST communication between MetaTrader and local services
• trading/risk-management logic

Limitations

• The repository is derived from article content and should be treated as a reference reconstruction unless the archive files are actually present.
• Publication date and article category were not available in the processed input.
• No performance claims beyond the article description should be assumed.
• The system depends on a local Flask server and MetaTrader 5 environment; deployment details may vary.
• The article describes code and file roles, but full verification of attached source completeness is not possible from the processed input alone.

Reference

Original MQL5 article: https://www.mql5.com/en/articles/22220