DRM-ARM-stats

DRM-ARM++

These are visualizations of decision trees trained on the probability that a cube has a “findable” sub-7 EO-to-DR solution.

The trees use the following features:

Feature Name Meaning
n_pairs Number of top pairs (bad corner + bad edge)
n_side_pairs Number of side pairs (white pair on F/B face)
n_fake_pairs Number of top fake-pairs (mis-oriented corner+edge)
corner_arm Number of out-of-AR corners
edge_arm Number of out-of-AR edges

At each level, the training algorithm chooses the feature that does the best job of separating the sub-7 solutions from the 7-plus solutions.

Trees by DRM:

4c4e corner case breakdown:

For a strategy on how to use these results in a solve, see DRM Subsets Meta