Genetic Programming Market Uptrend Predictor

🎯 Project Overview

This project uses Genetic Programming (GP) to discover early signals that predict market uptrends before they actually happen. Think of it like teaching a computer to evolve its own trading strategies by learning from historical market patterns.

The Core Idea

• Goal: Predict when EMA 12 will cross above EMA 50 (indicating an uptrend) before it actually happens
• Method: Use genetic programming to evolve mathematical expressions that can spot early warning signs
• Why: If we can detect uptrends 5-10 days early, we could potentially enter positions before the crowd notices

Simple Example

Instead of waiting for this to happen:

Day 10: EMA 12 crosses above EMA 50 → Everyone buys → Price already moved up

We want to detect patterns like:

Day 5: Our evolved algorithm spots unusual volume + price patterns
Day 6: Algorithm says "uptrend coming soon"
Day 10: EMA 12 crosses above EMA 50 → We're already positioned

🧬 How Genetic Programming Works Here

1. Population: Start with random mathematical expressions (like DNA)
2. Evolution: The best-performing expressions "breed" to create new ones
3. Selection: Expressions that correctly predict uptrends survive
4. Mutation: Small random changes keep exploring new patterns
5. Repeat: Over generations, we evolve better prediction formulas

📊 Target Definition

Uptrend Signal: When EMA 12 > EMA 50

• EMA = Exponential Moving Average
• This is a common technical indicator traders use
• We want to predict this crossover 3-10 days before it happens

📁 Project Structure

Core Files

• main.py: Entry point that orchestrates the entire GP workflow - downloads data, cleans it, labels it, runs evolution for 200 generations with population of 3000

• DataDownloader.py: Downloads market data from Yahoo Finance with support for chunked hourly data retrieval to work around API limitations

• DataCleaning.py: Cleans and validates market data by checking OHLC relationships, removing outliers, and handling missing values

• DataLabel.py: Labels data with prediction targets by identifying EMA crossovers and marking signals 3-10 days before they occur

• GPFramework.py: High-level orchestrator that provides a clean API to initialize and configure the entire GP framework

GP Framework Components (GP_Framework/)

• FitnessEvaluator.py: Evaluates GP individuals using F1 score with early detection bonus - rewards accurate predictions that come days before the actual crossover

• GeneticOperators.py: Defines genetic operators (crossover, mutation, selection) that create new individuals during evolution

• PopulationManager.py: Manages GP population creation and initialization using DEAP framework

• PrimitiveSetBuilder.py: Builds the primitive set defining all available functions and terminals for constructing trading signals

Technical Indicators (GP_Indicators/)

• BasicIndicators.py: Standard technical indicators (EMA, RSI, MACD, Bollinger Bands, ATR, Momentum) wrapped for GP use

• MomentumIndicators.py: Calculates momentum across multiple timeframes (hourly, daily, weekly) and their alignment

• MultiTimeframe.py: Calculates indicators on different timeframes to enable multi-timeframe analysis

• PositionEncoding.py: Encodes price position relative to timeframe ranges and moving averages for context

GP Primitives (GP_Primitives/)

• ComparisonPrimitives.py: Comparison and logical operators (>, <, AND, OR, IF-THEN-ELSE) for conditional logic

• MathPrimitives.py: Basic mathematical operations (protected division, min/max, sqrt, log, trig functions)

• TradingPrimitives.py: Trading-specific operations like crossovers, lags, and price changes for pattern detection