API Reference / datasets / generators /

clustering/

Synthetic clustering datasets.

Unlabelled data whose true grouping you know, which is the only way to check what a clusterer actually found. The shapes are chosen to separate algorithms that assume round clusters from those that do not.

Generators

  • Blobs: Isotropic Gaussian clusters. The easy case, and what k-means
assumes — every algorithm should handle it.
  • Moons: Two interleaved crescents. Not linearly separable, so k-means
splits them the wrong way while a density-based clusterer gets them right.
  • Circles: One ring inside another. Same lesson, concentric.
  • SwissRoll: A 2-D sheet rolled through 3-D. Points near each other in
space can be far apart along the sheet, which is the problem manifold learning exists to solve.

Notes

These return y as well: the true cluster assignment. Do not fit on it — it is there so you can score the result with a metric like adjusted_rand_score.
python 
>>> from tuiml.datasets.generators.clustering import Moons
>>> data = Moons(n_samples=200, noise=0.05, random_state=0).generate()
>>> data.X.shape
(200, 2)

Modules

blobs

Blobs generator: isotropic Gaussian clusters....

1 class · 0 functions
circles

Circles generator: two concentric rings....

1 class · 0 functions
moons

Moons generator: two interleaving half circles....

1 class · 0 functions
swiss_roll

Swiss roll generator: a 2-D sheet rolled through 3-D space....

1 class · 0 functions