Hierachical Clustering Bug/Feature

[izaychik63]

I’m using Hierachical Clustering node just behind Distance Matrix Calculate node. because different number of lines in my consecutive runs I’m getting a message that some measures are not here. Is it possible to add always include all columns option to avoid necessity to open and resubmit input columns for every run. At least k-mean and k-medoids
do not require to reopen node for consecutive runs.

[ana_ved]

Hi @izaychick63

Could you post a screenshot of what you get?

That is a good point, the hierarchical cluster node does not contain the include all columns options. I tried out controlling the included columns with a string list flow variable and I could overcome the behaviour. Maybe this could help you too

Cheers
Ana

[izaychik63]

Thank you, @ana_ved . I believe it is easy to add/fix. It seems that clustering is neglected to some extent. Say, k-medoids generate an error if number of clusters is too high; OPTICS generates an error and has no graphical output with some parameters combination. Happily k-means and DBSCAN are better done.

[izaychik63]

Ana, could you pleas post your example of headers variable use. I’m getting no columns …error.

Thank you
Igor

[ipazin]

Hi @izaychik63,

check here how to make sure to always include all columns using Column Filter Widget node:

Joiner node in Loop

Hello there, to answer my own topic from a while ago. This is now possible with array flow variables which were introduced with KNIME version 4.1. Who is interested check out example on KNIME Hub. Doesn’t feature a loop but you’ll get a point Br, Ivan

Br,
Ivan

[izaychik63]

@ipazin, this sounds as simple not working solution. I have 2 columns Key one as string and weight as integer. Only second one is allowed to be on columns parameter variable as it is numerical.
Please advice how it could be set for parameters below

I need to add group column later and cluster by group in a loop.

[ana_ved]

Hey @izaychik63

It would help me a bit if you explain a bit the use case. My idea was exactly the same as ivan. Here is a quick example of how I achieved it:

Example Variable Columns with Cluster – ana_ved

[izaychik63]

Thank you, Ana. I connected filter directly to the Hierarchical Clustering node. Your example explains that 2 connections are necessary. My assumption was that node will decide automatically what columns (numerical one) to use.