Performance Characterization

This page defines a reproducible process for performance-oriented ta4j evaluation.

Objectives

• Compare execution pathways under the same workload assumptions
• Understand precision/speed tradeoffs (DecimalNum vs DoubleNum)
• Capture stable baseline runs for regression monitoring

Reproducible commands

From repository root:

mvn -pl ta4j-examples exec:java -Dexec.mainClass=ta4jexamples.backtesting.BacktestPerformanceTuningHarness

Optional focused benchmark lane:

mvn -pl ta4j-examples test -Dtest=CachedIndicatorBenchmark -Dta4j.runBenchmarks=true

Interpretation checklist

• Hold data range and strategy set constant across runs
• Change one performance variable at a time (execution model, number type, candidate count)
• Record elapsed time, memory profile, and ranking stability
• Treat major speedups with skepticism if strategy ordering changes unexpectedly

Number-type guidance

• Prefer DecimalNum for precision-sensitive or audit-heavy analysis
• Prefer DoubleNum when throughput dominates and floating-point tradeoffs are acceptable

Related pages

• Num
• Backtesting
• Execution Decision Matrix