Bug: min_cluster_samples should not be set to a non-integer

[michaelkerber]

Is your feature request related to a problem? Please describe.
KeplerMapper does not work in combination with sklearn.clustering.AgglomerativeClustering (using distance_threshold). It seems that KeplerMapper relies on the parameter "min_clusters/n_clusters/min_samples" to be available (kmapper.py, line 506).

Describe the solution you'd like
The documentation should be more clear about what is expected from the clustering method to work in KeplerMapper.

[deargle]

That code block has a final fallback of 2 if none of the previous are available:

kepler-mapper/kmapper/kmapper.py

Lines 506 to 511 in 56dbd66

	min_cluster_samples = cluster_params.get(
	"n_clusters",
	cluster_params.get(
	"min_cluster_size", cluster_params.get("min_samples", 2)
	),
	)

[deargle]

Also, here's an example of someone using AgglomerativeClustering #185 (comment) , what error message are you getting? Full stack trace please?

[michaelkerber]

I tried this clustering method:
cluster_method=sklearn.cluster.AgglomerativeClustering(n_clusters=None,distance_threshold=0.2,compute_full_tree=True,linkage='single')

Traceback (most recent call last):
File "mapper_cat.py", line 32, in
cover=my_cover,
File "/home/mkerber/.local/lib/python3.7/site-packages/kmapper/kmapper.py", line 529, in map
if hypercube.shape[0] >= min_cluster_samples:
TypeError: '>=' not supported between instances of 'int' and 'NoneType'

I think the error makes sense, since n_clusters is set, just not to an integer. The documentation of AgglomerativeClustering says that n_clusters has to be set to None if distance_threshold is set.

Perhaps you can check whether the parameter is an integer and if not, set it to 2 instead?

[deargle]

Indeed! Thanks for the bug report.I remember now that @torlarse had hard-coded kmapper to use min_cluster_samples = 2 when they posted the code in that comment. Maybe you can try that until a patch is released?