- glass.txt.gz, (2.4 KB).
- hepat.data.gz, (2.6 KB.
- ionosphere.data.gz, (28 KB).
- iris_all.txt.gz, (757 bytes).
- liver.data.gz, (3.0 KB).
- nvowel.data.gz, (14 KB).
- oq.data.gz, (17 KB).
- uv.data.gz, (17 KB).
- vote.data.gz, (1.2 KB).

From *The Elements of Statistical Learning*, http://http://www-stat.stanford.edu/~tibs/ElemStatLearn

- 14cancer.xtrain_transposed.gz, (4.5 MB).

Synthetic 2D data.

- 2dmulti_1200.txt.gz, (15 KB). The 2D synthetic data set was generated from eight distributions: a zero mean Gaussian with unit covariance matrix, a Gaussian centered at (0,-6) with a covariance matrix created by scaling the minor axis scale to 1/8 of the major axis and rotated by $\frac{\pi}{6}$. The circles at centered at (6,-5) with radii of 0.5, 1.25, and 2.0 respectively. The spirals are centered at (6,0) with scaling $\frac{1}{2.75\pi}$. The circles and spirals were sampled at uniformly over arc length and corruptive with Gaussian noise with $\sigma=0.1$ for the circles and $\sigma=0.02$ for the spirals. The last distribution is uniform noise over the range of the graph. The relative weighting of each distribution is 0.15 for each Gaussian, 0.3 total for all circles, 0.35 total for both spirals, and 0.05 for the uniform background noise. Generated from python scripts : genFig2.py, spiralgen.py, circlegen.py, and gaussgen.py.