Visualizing PMML Model (k-Means)

KNIME Analytics Platform
1

Hi,

is there any node or methode to visualize a PMML Cluster model (output of k-means node) ?

Thanks,

Michel

2

Hi Michel,
It depends a bit on what you would like to see. There is not a single node, but different techniques. If your data has 2 dimensions, you can simply add a Color Appender and assign a color based on the cluster to each row, then use a Scatter Plot node to see how the clusters are distributed. If you have 3 dimensions, you can do the same with the 2D/3D Scatterplot node, but it will only be really useful in an interactive scenario, not so much in a static report.
Alternatively, for 3 or more dimensions, you can try to reduce the dimensionality of your dataset with the PCA node and then visualize it with a scatterplot. Of course some information will be lost by the dimensionality reduction, but you can still get a glimpse at the data and cluster assignment.
Kind regards,
Alexander

2 Likes
3

Hi Alexander,

thanks for your suggestions. Actually I want to show mor than 3 dimensions, about 20. The main purpose to get impulse for further analyse.

I could not found any “visualizing” for clusters with more then 3 diemsnions. I hoped, over PMML it would be possible.

I’m going to use for now Excel and later I’ll write some code in Python to get the desired representation of clusters.

I was surprised regarding the missing represantation. I can rememeber, “my” very first tool 20 years ago (CA Clever Path Analysis Server) did it.

Warm regards,

Michel

4

Hi Michel,
Can you show a screenshot what visualization you expect? We can extract info from the PMML document and visualize it, but I am not sure what you are looking for.
Kind regards,
Alexander

5

Hello Alexander,

I only have a tree view, (right 20 years old :grinning: )

image
image842Ă—399 51.9 KB

but I would like to have a table to be able “compare” the individual features with each other.

In the cells of the Excel sheet I currently only have the average values, I would like more:

Min, max, …

And also cluster attributes such as

Centroid, Radius,…

I also have ideas for the visualizations, but I won’t show them until it has reached a certain level of maturity.

Warm regards,

Michel

6

Hi Michel,
This information is not contained in the PMML document anyways. But you can easily create it yourself from the table KNIME outputs from the k-Means node. Just add a GroupBy node, group by the assigned cluster and aggregate the other attributes (mean, max, min, …). You can then also add a Math Formula node to calculate the radii from the min and max values, if you want.
Kind regards,
Alexander

2 Likes
7

Hi Alexander,

thanks, already did it :slight_smile:

Warm regards,

Michel

8

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.