Skip to content
TradingTune

Bisection (Smart Search)

A fast 3-phase search: coarse sweep, golden-section narrow, top-value polish.

Quality
High
Speed
Fast
Budget
O(N)
Add to Chrome, it's free

How it works

Bisection works the way you would tune a dial by ear. First a coarse sweep checks a handful of values across each parameter's range to find roughly where the good region is.

Then it narrows: it repeatedly halves the bracket around the best value, keeping the better half each time, until the window is one step wide. Finally a polish phase tries combinations of the top values from each parameter, so it catches cases where two settings only shine together.

It is deterministic and resumable: the same inputs always produce the same plan, and you can stop and continue without losing your place.

Under the hood

Phase one takes about five step-aligned samples per numeric parameter (and both values of each boolean), one parameter at a time, to detect multi-modality before committing to a region. It picks the contiguous window with the best average metric, not the single best point, so a lucky spike next to a cliff does not mislead it.

Phase two is golden-section bisection inside that window, costing roughly log2(span / step) samples per parameter. Phase three is a combinatorial polish over the top two values per parameter, de-duplicated against everything already tried and capped at 256 combinations.

Each parameter is assumed to have one main peak after the coarse sweep. Truly noisy or strongly coupled landscapes can settle on a local maximum, which is exactly when you want the hybrid that adds a Bayesian polish.

Coarse samples
~5 per numeric
Narrowing
Golden-section
to within one step
Polish cap
256 combinations

When to use it

Reach for it when

  • Smooth-ish landscapes with mostly independent numeric parameters
  • When you want a fast, deterministic first pass
  • When repeatability and resumability matter

Reach for something else when

  • Rugged or noisy surfaces with many local optima (prefer TPE)
  • Strongly coupled parameters that only matter together

Want to see optimization in action? Browse our real strategy backtests vs buy and hold on NVDA, TSLA, and AAPL.

New to this? Read how to run your first optimization or learn how to choose the right optimization method for your strategy.

Related methods