tradestream-ga

Genetic Algorithm engine for TradeStream — strategy parameter optimization using evolutionary algorithms.

Extracted from tradestreamhq/tradestream.

Architecture

This repo contains the Jenetics-based GA engine that discovers optimal trading strategy parameters. It calls tradestream-backtest via gRPC as its fitness evaluator.

Tech Stack

Component	Technology
Language	Kotlin / Java 21
GA Engine	Jenetics 7.2
Pipeline	Apache Beam 2.62 on Flink
Build	Gradle (Kotlin DSL)
Messaging	Apache Kafka
Storage	PostgreSQL
Transport	gRPC / Protocol Buffers

gRPC Interface

GAService

Defined in src/main/proto/protos/ga_service.proto.

RPC	Description
SubmitStrategySpec	Kick off a GA optimization run for a given strategy/symbol/timeframe
GetTopStrategies	Retrieve top N strategies from a completed or in-progress run
GetPerformance	Get detailed performance metrics for a specific strategy

Data Flow

  Kafka (requests)          tradestream-backtest
       │                         ▲
       ▼                         │ gRPC: RunBacktest
  ┌─────────────┐           ┌────┴──────┐
  │ Discovery   │──────────►│ Fitness   │
  │ Pipeline    │           │ Function  │
  │ (Beam)      │◄──────────│ (remote)  │
  └──────┬──────┘           └───────────┘
         │
    ┌────┴────┐
    ▼         ▼
  Kafka    PostgreSQL
  (results) (persist)

Inputs

• Discovery requests via Kafka or gRPC (StrategyDiscoveryRequest)
• GA parameters: population size, max generations, mutation/crossover rates

Outputs

• Discovered strategies: Top N optimized parameter sets with fitness scores
• Run status: SUBMITTED → RUNNING → COMPLETED

Building

./gradlew build

Running

Requires tradestream-backtest to be running on the configured gRPC endpoint.

./gradlew run

Relationship to tradestream-backtest

This repo does not contain TA4J or any strategy implementation code. All fitness evaluation is delegated to tradestream-backtest via gRPC. This separation allows:

• Independent scaling of GA compute vs. backtesting compute
• Swapping the backtesting engine without changing the GA
• Running multiple GA instances against a shared backtesting service