Benchmark Datasets

The toolkit ships deterministic synthetic fixtures so examples, tests, and benchmark smoke reports use the same topology contracts.

Public API

import topoml

names = topoml.list_benchmark_datasets()
circle = topoml.make_noisy_circle(n_samples=64, noise=0.0, random_state=7)
two = topoml.load_benchmark_dataset("two_circles", n_samples=32)
bridge = topoml.make_cluster_bridge()

Each call returns BenchmarkDataset:

Field	Meaning
name	Stable fixture name
points	NumPy point cloud
labels	Labels for ML smoke tests or visualization
expected_betti	Topology contract such as beta1 = 1
metadata	Generation settings and fixture notes

Fixtures

Fixture	Purpose	expected_betti
make_noisy_circle	One loop for H1 tutorials and PH regressions	{"beta0": 1, "beta1": 1}
make_two_circles	Two components and two loops for clustering and Mapper demos	{"components": 2, "loops": 2}
make_cluster_bridge	Three-point H0 merge fixture used by the E2E claim gate	{"beta0@0.1": 3, "beta0@0.3": 2, "beta0@6": 1}

flowchart LR
  A["Dataset fixture"] --> B["Expected topology metadata"]
  B --> C["Persistent homology smoke test"]
  C --> D["Tutorial or benchmark artifact"]
  D --> E["Published claim boundary"]

Claim Boundary

These are synthetic benchmark fixtures. They prove that examples and tests share known topology contracts. They do not replace real domain datasets, leaderboard benchmarks, or task-specific validation.