learn_lisa_clusters

learn_lisa_clusters(
    side=12,
    block=3,
    shift=2.0,
    alpha=0.05,
    permutations=999,
    seed=0,
)

Plant hot and cold spots, then watch LISA find them (and sometimes cry wolf).

Draws iid noise on a side x side lattice, shifts a block x block hot block up by shift and a cold block down by the same amount, and runs local Moran (LISA) with significance masking at alpha. The map shows the recovered High-High / Low-Low clusters with the planted blocks outlined; the summary reports the hit rates and the false-positive share — a reminder that with hundreds of local tests, some cells are flagged by chance alone.

Parameters

Name	Type	Description	Default
side	int	Lattice side length (n = side²); must fit two disjoint blocks.	12
block	int	Side length of each planted block.	3
shift	float	Size of the planted level shift (in standard deviations of the noise).	2.0
alpha	float	Significance level masking the cluster labels (p_sim < alpha).	0.05
permutations	int	Conditional permutations behind the local pseudo p-values.	999
seed	int	Random seed (drives both the noise and the permutations).	0

Returns

Name	Type	Description
	SandboxResult	df (one row per cell with planted and detected labels), fig, summary, topic and the raw field in data.

Examples

import geometrics as gm

res = gm.learn_lisa_clusters(shift=2.0)
print(res.summary["sensitivity_hot"], res.summary["false_positive_rate"])