Objective metrics

Objective metrics tell HyperOptimizer what “best” means. You can emit many metrics; the experiment configuration decides which ones are optimized.

One or more objectives

An experiment can define multiple maximize/minimize objectives. Metric names must match the keys you print as hpo.metrics.<key>.

If two metrics matter but you only want to optimize one, keep a single objective and treat the others as guardrails when reviewing results.

Maximize

Use for metrics where larger is better, such as sharpe, profit_factor, or accuracy.

Minimize

Use for metrics where smaller is better, such as loss, latency_ms, or max_drawdown.

Guardrail

Emit secondary metrics to avoid configurations that look good on one number but fail operationally.

Emit supporting metrics

Supporting metrics make the dashboard useful even when they are not objectives.

metrics = {
    "objective": score,
    "sharpe": sharpe,
    "max_drawdown": drawdown,
    "total_trades": total_trades,
}

Trading examples

For trading strategy optimization, objective choice matters. A high-profit configuration with huge drawdown may be worse than a steadier configuration.

Metric	Role	Notes
`sharpe`	maximize	Useful when risk-adjusted returns matter more than raw profit.
`max_drawdown`	minimize	Useful as a guardrail for downside risk.
`profit_factor`	maximize	Useful for comparing gross profit against gross loss.
`total_trades`	guardrail	Helps avoid overfitting on too few trades.