raystrack 1.0.2

Lightweight Monte-Carlo view-factor solver with CPU, CUDA and BVH paths

Raystrack

Lightweight Monte-Carlo view-factor solver for polygonal meshes.

Raystrack computes radiative view factors F(i->j) between triangulated surfaces using quasi-Monte-Carlo ray tracing. It runs on CPU, can leverage Numba/CUDA on NVIDIA GPUs when available, and optionally accelerates ray intersection with a BVH. The repository also ships a pure-Python API you can use outside Rhino.

Features

• Efficient Monte-Carlo view factors: front/back hits, optional reciprocity
• CPU and optional CUDA GPU backends (Numba)
• Optional BVH acceleration structures
• Python API plus Grasshopper components (Rhino installer coming soon)

Installation

Python package

Use Raystrack as a normal Python package outside Rhino or Grasshopper.

From a local clone of this repository:

pip install .

Or from an absolute path:

pip install /path/to/raystrack

Requirements: Python 3.9+, numpy, numba. CUDA acceleration is enabled automatically when numba.cuda detects a compatible GPU.

Rhino / Grasshopper status

The Rhino 8 Grasshopper installers are not yet published. Follow this README or the issue tracker for updates when a tested Rhino workflow becomes available.

Examples

All examples live in examples/. Start by running ex00_street_canyon_geometry to generate street_canyon.json; subsequent scripts expect that file. Each Python example now documents its own inputs and parameters inline, so open the scripts directly for usage guidance and tunable options.

Quick start (Python)

import numpy as np
from raystrack import view_factor_matrix, MatrixParams

# Each mesh is a tuple: (name: str, V: (N,3) float32, F: (M,3) int32)
V_a = np.array([[0, 0, 0], [1, 0, 0], [1, 1, 0], [0, 1, 0]], dtype=np.float32)
F_a = np.array([[0, 1, 2], [0, 2, 3]], dtype=np.int32)  # two triangles

V_b = np.array([[0, 0, 1], [1, 0, 1], [1, 1, 1], [0, 1, 1]], dtype=np.float32)
F_b = F_a.copy()

meshes = [
    ("A", V_a, F_a),
    ("B", V_b, F_b),
]

params = MatrixParams(
    samples=256,   # sampling density per unit area (QMC grid)
    rays=256,      # rays per cell
    bvh="builtin", # optional BVH acceleration (auto|off|builtin)
    reciprocity=True,
)

res = view_factor_matrix(meshes, params=params)

print(res["A"])  # e.g. {"B_front": 0.5, "B_back": 0.0, ...}

To persist a matrix without directional suffixes in the saved JSON:

from raystrack import save_vf_matrix_json

save_vf_matrix_json(res, "vf_matrix.json", strip_dir=True)

This collapses receiver keys like "B_front" and "B_back" into "B" and sums both values per sender row.

Parameter presets

Raystrack uses two parameter containers to keep configuration consistent:

• MatrixParams: controls the scene-to-scene view-factor solve (sampling, BVH, device selection, convergence tolerances, and reciprocity enforcement).
• SkyParams: controls the sky view-factor solve (sampling, device selection, and convergence tolerances, plus discrete=True for 145 sky patches or False for a single merged "Sky" output).

Typical usage:

from raystrack import MatrixParams, SkyParams, view_factor_matrix, view_factor_to_tregenza_sky

matrix_params = MatrixParams(samples=32, rays=256, reciprocity=True, flip_faces=False)
sky_params = SkyParams(samples=32, rays=256)

vf_scene = view_factor_matrix(meshes, params=matrix_params)
vf_sky = view_factor_to_tregenza_sky(meshes, params=sky_params)

License

MIT - see LICENSE.