How alanmac07/trading-agent works
This repository combines an evolutionary trading simulator and dashboard with a separate NEAT training and holdout-evaluation workflow. In the interactive path, market data feeds the FastAPI server and evolution engine; in the research path, data is engineered and split, then evolved across episodes and scored. The README also describes a legacy continuous-engine feature pipeline, but its agent and dashboard files are absent from this tree. Some frontend-to-server wiring is not established by the excerpts, so that link is marked unverified.
Main entry points: build_dataset.py Dataset builder · trainer.py Population trainer · evaluate_holdout.py Holdout evaluator · price_data_multi.csv Historical market data
Components 20
Interactive simulation
- Simulation API
server.py - Evolution engine
evolution.py
NEAT research engine
- Dataset builder
neat_engine/build_dataset.py - NEAT features
neat_engine/features.py - Episode windows
neat_engine/episodes.py - Population trainer
neat_engine/trainer.py - Fitness evaluation
neat_engine/fitness.py - Genome network
neat_engine/network_agent.py - Trade risk rules
neat_engine/risk_manager.py - Best genome
results - Holdout evaluator
neat_engine/evaluate_holdout.py
Market data
- Historical market data
data/price_data_multi.csv - Market data fetcher
fetch_data.py - Continuous indicators
feature_engineering.py - Training datasets
data - Holdout datasets
data
Dashboard interface
- Dashboard page
frontend/src/app/page.tsx - Agent views and charts
frontend/src/components - API client
frontend/src/lib/api.ts
Other components
- User
Connections 22
- User → Dashboard page: uses
README.md - Dashboard page → API client: requests data
- API client → Simulation API: calls API
- Simulation API → Evolution engine: runs simulation
server.py - Simulation API → Historical market data: loads data
server.py - Market data fetcher → Historical market data: writes prices
README.md - Continuous indicators → Historical market data: reads prices
README.md - Dataset builder → Historical market data: reads prices
neat_engine/build_dataset.py - Dataset builder → NEAT features: computes features
neat_engine/build_dataset.py - Dataset builder → Training datasets: writes split
neat_engine/build_dataset.py - Dataset builder → Holdout datasets: writes split
neat_engine/build_dataset.py - Population trainer → Episode windows: builds windows
neat_engine/trainer.py - Population trainer → Fitness evaluation: scores genomes
neat_engine/trainer.py - Population trainer → Training datasets: reads training data
README.md - Fitness evaluation → Genome network: queries signals
neat_engine/fitness.py - Fitness evaluation → Trade risk rules: gets trade decisions
neat_engine/fitness.py - Population trainer → Best genome: saves winner
neat_engine/trainer.py - Holdout evaluator → Genome network: queries signals
neat_engine/evaluate_holdout.py - Holdout evaluator → Trade risk rules: gets trade decisions
neat_engine/evaluate_holdout.py - Holdout evaluator → Holdout datasets: reads holdout
neat_engine/evaluate_holdout.py - Holdout evaluator → Best genome: loads winner
neat_engine/evaluate_holdout.py - Dashboard page → Agent views and charts: renders views