ES

Performance Tips

Heuristics

SheShe applies simple rules to keep runtimes manageable. The primary one is auto_rays_by_dim, which caps base_2d_rays as dimensionality grows:

• 2–24 features → use the full base_2d_rays value (default 32).
• 25–64 features → cap base_2d_rays at 16.
• 65+ features → cap base_2d_rays at 12.

Disable this behaviour with auto_rays_by_dim=False to keep the original ray count.

High-dimensional datasets

Large feature spaces require additional care. Consider the following adjustments when dealing with tens or hundreds of dimensions:

• Reduce rays: set base_2d_rays to 16 or 12 when auto_rays_by_dim is disabled or a lower cap is needed.
• Trim scan steps: drop scan_steps from 24 to 12 to halve the evaluations along each ray.
• Limit exploration: decrease n_max_seeds and max_subspaces to restrict the number of starting points and subspaces scanned.
• Preselect subspaces: run SubspaceScout to focus on promising feature pairs before launching heavier searches.

See the README for additional context on how these parameters interact.