"""Auto-generated example: Adaptive detector — tuning noise."""
import random
import numpy as np
import bencher as bn
from bencher.regression import detect_adaptive, render_regression_png
def _render_detection_png(hist, current, result):
"""Render the adaptive-detector outcome as a PNG and return its path."""
return render_regression_png(
result, hist, current,
path=bn.gen_image_path(f"regression_{result.method}"),
figsize=(4.5, 3.2), dpi=100,
)
_REGRESSION_STEP = 25.0
class AdaptiveNoiseRobustness(bn.ParametrizedSweep):
"""Fixed 25-unit regression with varying noise and regression_percentage."""
noise_sigma = bn.FloatSweep(
default=10.0, bounds=[0.0, 40.0],
doc="Noise standard deviation (0.0 collapses MAD and leaves only the "
"percentage band).",
)
regression_percentage = bn.FloatSweep(
default=10.0, bounds=[0.0, 40.0],
doc="Minimum percent change required to flag a regression (dual-band "
"AND gate). 0.0 disables the percentage gate.",
)
detection_plot = bn.ResultImage(doc="Regression diagnostic PNG")
_Z_THRESHOLD = 3.5
_N_HIST = 20
_N_REPEATS = 5
def benchmark(self):
baseline = 100.0
hist_2d = np.array([
[baseline + random.gauss(0, self.noise_sigma) for _ in range(self._N_REPEATS)]
for _ in range(self._N_HIST)
])
hist_means = hist_2d.mean(axis=1)
current = np.array(
[baseline + _REGRESSION_STEP + random.gauss(0, self.noise_sigma)
for _ in range(5)]
)
pct = self.regression_percentage if self.regression_percentage > 0.0 else None
result = detect_adaptive(
"metric", hist_means, current,
regression_mad=self._Z_THRESHOLD,
direction=bn.OptDir.minimize,
regression_percentage=pct,
historical_samples=hist_2d.ravel(),
)
# git_time_event-style string labels for the x-axis; historical_all_x
# stays numeric (integer positions aligned with the tick labels) so
# the per-sample scatter still renders on the categorical axis.
result.historical_x = np.array(
[f"2024-01-{i + 1:02d} v{i:02d}" for i in range(self._N_HIST)]
)
result.current_x = f"2024-01-{self._N_HIST + 1:02d} v{self._N_HIST:02d}"
result.historical_all = hist_2d.ravel()
result.historical_all_x = np.repeat(np.arange(self._N_HIST), self._N_REPEATS)
self.detection_plot = _render_detection_png(hist_means, current, result)
def example_regression_tuning_noise(run_cfg: bn.BenchRunCfg | None = None) -> bn.Bench:
"""Adaptive detector — tuning noise."""
bench = AdaptiveNoiseRobustness().to_bench(run_cfg)
bench.plot_sweep(
input_vars=['noise_sigma', 'regression_percentage'],
result_vars=["detection_plot"],
description='A fixed +25 step regression is present, but the noise level varies. At low noise the regression is obvious and the MAD acceptance band is tight; at high noise the signal is buried and the MAD band is wide. The noise_sigma=0 row is the pathological case: MAD collapses to zero, the MAD band is a hairline, and only the percentage band is visible — that row shows the percentage acceptance band on its own. Increase regression_percentage in any row to see the percentage band expand and AND-gate the verdict.',
)
return bench
if __name__ == "__main__":
bn.run(example_regression_tuning_noise, subsampling_divisions=3)