Dave Eargle

> By using affinity=precomputed, doesn't Scikit require distance_treshold to be set? Which scikit modeler are you referring to? DBSCAN doesn't -- I have used DBSCAN thusly: `clusterer=km.cluster.DBSCAN(eps=.5, min_samples=4, metric='precomputed')` >...

I re-read your earlier comment: > To my understanding this defines the minimal number of samples in each cube of the covering. This certainly works as a reduction method but...

Perhaps the default value of `n_clusters` for the [constructor](https://github.com/scikit-learn/scikit-learn/blob/e5698bde9a8b719514bf39e6e5d58f90cfe5bc01/sklearn/cluster/_agglomerative.py#L761) could be used as a fallback. For agglomerativeclustering, that's `2`. [This](https://stackoverflow.com/a/12627202/5917194) method can be used to get a function's default argument...

Or maybe if cluster_params `distance_threshold` is set, then default to min `2`. I don't know how well that would generalize across scikit clusterers though.

@sauln what do you think about the following: allow `min_cluster_samples` to be passed to `.map()`, which would * be preferred over any cluster_params in determining whether or not to skip...

I think agglomerative clustering is common enough that ideally kmapper should accommodate it without users having to modify kmapper.py -- glad you found a workaround though. On Tue, Jan 7,...

@sauln I'm going to make the final fallback be to min of 2, not min of 1, and take out the hard fix of 2 for precomputed matrices

I just found this again because I have @torlarse 's original problem -- I want to filter out nodes that are too small for my liking, and `min_samples` doesn't do...

What behavior do you desire? That it stay like it was before the refresh? Any "saved" node positioning is technically possible using browser local storage. The positioning is done via...

Not possible at the moment, but I have tinkered with better control over color functions before, and I can implement something that lets a user switch between several color functions...