CloudDelta.Benchmark (cloud_delta v0.2.0)

Dataset generation and honest compression benchmarks.

Every reported ratio is raw_float32_bytes / encoded_byte_size of a real binary. Reconstruction is checked on that same binary.

Summary

Functions

generate_dataset(n \\ 5, pattern \\ :squared)

Generate a synthetic 2D point cloud.

generate_points_3d(n, pattern \\ :clustered)

Generate a synthetic 3D point list ({x, y, z} tuples).

measure(n, pattern, opts \\ [mode: :quantized, bits: 16])

Measure one dataset with a real compress/uncompress round-trip.

run_benchmark_suite(patterns \\ [:random, :linear, :squared, :clustered, :grid], sizes \\ [1000, 10000], opts \\ [mode: :quantized, bits: 16])

Print an honest benchmark table.

Functions

generate_dataset(n \\ 5, pattern \\ :squared)

Generate a synthetic 2D point cloud.

Patterns:

:squared — y = x² + noise
:sin — shifted sine plus noise
:linear — y = 2x + noise
:random — uncorrelated uniform noise
:clustered — five tight Gaussian blobs (spatially coherent)
:grid — regular lattice plus tiny noise

generate_points_3d(n, pattern \\ :clustered)

Generate a synthetic 3D point list ({x, y, z} tuples).

measure(n, pattern, opts \\ [mode: :quantized, bits: 16])

Measure one dataset with a real compress/uncompress round-trip.

run_benchmark_suite(patterns \\ [:random, :linear, :squared, :clustered, :grid], sizes \\ [1000, 10000], opts \\ [mode: :quantized, bits: 16])

Print an honest benchmark table.