Size Scaling Example

This example generates spherical meshes of increasing resolution and reports the cost of a small set of core PyMembrane operations on each mesh.

What this example demonstrates

• sphere generation from a recursive icosphere construction

• energy evaluation only

• Monte Carlo vertex-displacement attempts

• Monte Carlo edge-flip attempts

• Brownian-dynamics vertex updates

• optional output timing, reported separately when --dump-output is used

The benchmark also records environment information such as the CPU model, the operating system, the Python version, the PyMembrane version or git commit, and the number of threads when that information is available.

How to run

python -m pymembrane.examples.size_scaling --quick
python -m pymembrane.examples.size_scaling --vertices 1000 5000 10000 50000 --steps 10000 --repeat 3 --json size_scaling.json

If you want to keep the generated temporary mesh files on disk, add --keep-meshes and --output-dir:

python -m pymembrane.examples.size_scaling --keep-meshes --output-dir size_scaling_meshes

Command-line arguments

Argument	Default	Description
--quick	False	Run a short version suitable for checking the installation and the benchmark workflow.
--subdivisions	None	Icosphere subdivision levels used to generate the benchmark meshes. Runtime default: [0, 1, 2] with --quick or [0, 1, 2, 3] otherwise. This option is mutually exclusive with --vertices.
--vertices	None	Target vertex counts. Each requested size is mapped to the nearest available icosphere subdivision level. This option is mutually exclusive with --subdivisions.
--steps	None	Number of evolution steps or repeated energy evaluations used in each benchmark row. Runtime default: 100 with --quick or 1000 otherwise.
--repeat	None	Number of timed repeats collected for each benchmark row. Runtime default: 3.
--warmup	None	Number of untimed warmup runs collected before timing each benchmark row. Runtime default: 1.
--json	None	Write the timing summary to a JSON file.
--csv	None	Write the timing summary to a CSV file.
--dump-output	False	Time a representative VTK output write in addition to the compute benchmarks.
--keep-meshes	False	Keep the generated temporary mesh files on disk.
--output-dir	size_scaling_meshes	Directory used when keeping generated mesh files.

Expected output

The example prints a compact table with rows for energy_eval_only, mc_vertex_move, mc_edge_flip, brownian_dynamics, and, when requested, dump_vtk. If --json or --csv is supplied, the results are written to the chosen file.

Quick mode

--quick reduces the default subdivision list to 0 1 2 and uses a short run of 100 steps, --repeat 3, and --warmup 1. For more stable timings, increase --steps and keep --repeat 3 or larger.