Herbie results

Time: 6.6min

Passed: 8/9

Tests: 25

Bits: 95/103

Output vs Input Accuracy

Each point represents a Herbie run below. Its horizontal position shows initial accuracy, and vertical position shows final accuracy. Points above the line are improved by Herbie.

Accuracy vs Cost

A joint cost-accuracy pareto curve for the Herbie runs below. Accuracy is on the vertical axis, and cost is on the vertical axis. Down and to the left is better. The initial programs are shown by the red square.

Test	Start	Result ?	Time
UniformSampleCone, z	0.0	0.0	5.9s	»
UniformSampleCone, y	13.8	0.5	18.3s	»
UniformSampleCone, x	13.8	0.3	20.3s	»
UniformSampleCone 2	0.3	0.3	23.3s	»
Trowbridge-Reitz Sample, sample surface normal, cosTheta	0.2	0.2	30.8s	»
Trowbridge-Reitz Sample, near normal, slope_y	0.5	0.5	12.4s	»
Trowbridge-Reitz Sample, near normal, slope_x	0.3	0.3	11.0s	»
Sample trimmed logistic on [-pi, pi]	0.3	0.3	15.2s	»
Logistic function	0.1	0.0	12.0s	»
Logistic distribution	0.1	0.2	14.0s	»
Lanczos kernel	0.7	0.7	15.5s	»
HairBSDF, sample_f, cosTheta	0.2	0.2	20.4s	»
HairBSDF, gamma for a refracted ray	2.4	0.5	17.5s	»
HairBSDF, Mp, upper	0.4	0.4	16.5s	»
HairBSDF, Mp, lower	0.1	0.1	15.0s	»
GTR1 distribution	0.5	0.4	16.2s	»
Disney BSSRDF, sample scattering profile, upper	1.3	0.5	13.7s	»
Disney BSSRDF, sample scattering profile, lower	12.5	0.2	11.8s	»
Disney BSSRDF, PDF of scattering profile	0.1	0.1	13.9s	»
Curve intersection, scale width based on ribbon orientation	0.9	0.6	19.7s	»
Beckmann Sample, normalization factor	0.7	0.4	16.2s	»
Beckmann Sample, near normal, slope_y	13.6	0.5	14.5s	»
Beckmann Sample, near normal, slope_x	13.6	0.3	14.1s	»
Beckmann Distribution sample, tan2theta, alphax == alphay	14.1	0.3	9.6s	»
Beckmann Distribution sample, tan2theta, alphax != alphay, u1 <= 0.5	12.6	0.5	18.1s	»