# Эксперимент: Entropy-calibrated hyperbolic curvature (#1226)

{
  "worked": true,
  "confidence": 7,
  "verdict": "Built a NumPy/SciPy MVP for the conditional farthest-triplet entropy master curve, monotone inversion, curvature amplification A(s)=s coth(s), and a slow EMA curvature controller. The core claim was numerically observed: entropy decreased strictly over the tested lambda grid (H(0)=1.0986, H(5)=0.2104), and the controller reduced mean absolute curvature error from 1.3667 for fixed initializations to 0.0204. Anisotropy produced inference errors up to 0.219, supporting the stated robustness limitation; the apparent controller win is synthetic and model-consistent rather than evidence from a trained neural embedding.",
  "metrics": {
    "baseline": "Fixed curvature at initial kappas [0.15, 1.0, 4.0]: mean absolute error to target kappa=1.3667.",
    "idea": "Entropy-inverted curvature EMA: final kappas [0.7539, 0.7569, 0.8004] for target 0.75; mean absolute error=0.0204. Entropy curve decreasing fraction=1.0, with H(lambda=0)=1.0986 and H(lambda=5)=0.2104. Under anisotropic angular scales [1.0,2.0,0.5], inferred-lambda absolute errors were 0.092, 0.019, and 0.219 at true lambda [0.5,1.5,3.0]."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 entropy_curvature.py",
  "files": [
    "entropy_curvature.py",
    "results.json"
  ],
  "limitations": "No trained Poincare-ball encoder, MNIST/CIFAR classification, validation accuracy, calibration, wall-clock comparison, or real hyperbolic-distance experiment was tested. The master curve assumes a standard-normal calibrated radial law and isotropic angular behavior; the controller target entropy is generated from the same assumed model, so the recovery result does not establish an end-to-end ML improvement."
}