ppforest2::cli::ConvergenceCriteria Struct Reference

Convergence criteria for adaptive stopping in benchmarks. More...

#include <Benchmark.hpp>

Public Attributes
float	cv = 0.05f
int	max = 200
int	min = 10
int	window = 3

Detailed Description

Convergence criteria for adaptive stopping in benchmarks.

Instead of running a fixed number of iterations, convergence mode monitors the coefficient of variation (CV = std/mean) of timing measurements and stops once results are statistically stable.

The algorithm works as follows:

1. After each iteration, compute CV across all timing samples.
2. If CV < cv_threshold, increment a stability counter. Otherwise, reset the counter to zero.
3. Stop when the counter reaches stable_window (i.e., CV stayed below threshold for stable_window consecutive iterations).
4. Never check before min_iterations (need enough samples).
5. Never exceed max_iterations (hard cap to prevent runaway).

Member Data Documentation

cv

float ppforest2::cli::ConvergenceCriteria::cv = 0.05f

Target CV threshold (e.g., 0.05 = stop when std < 5% of mean).

max

int ppforest2::cli::ConvergenceCriteria::max = 200

Hard upper bound on iterations (stops even if not converged).

min

int ppforest2::cli::ConvergenceCriteria::min = 10

Minimum iterations before convergence checks begin.

window

int ppforest2::cli::ConvergenceCriteria::window = 3

Number of consecutive iterations that must stay below the CV threshold.

---

The documentation for this struct was generated from the following file:

• core/src/cli/Benchmark.hpp