All Apps and Add-ons

Obtaining cluster centres details from K-Means algorithm

kiril123
Path Finder

I am using K-Means algorithm from Machine Learning toolkit to cluster some data.
After algorithm has converged i can see two new fields appended to the original data - cluster ID and cluster distance.

This is great, however I also need cluster centre details for each cluster. I need this information to calculate distance to each cluster centre from new data points and then assign these data points to the appropriate cluster.

Is there any way to accomplish this in Splunk?

0 Karma
1 Solution

aoliner_splunk
Splunk Employee
Splunk Employee

You can inspect the KMeans model you built with fit using the summary command:

| summary <your_model_name>

Although, if you're trying to assign new points to the appropriate cluster, you can simply apply your model:

<new_points> | apply <your_model_name>

View solution in original post

aljohnson_splun
Splunk Employee
Splunk Employee

You can also use the cluster distance to calculate centroids of a sort:

| inputlookup iris.csv 
| fit KMeans k=3 petal* 
| eval point_size = 1 
| appendpipe 
    [| stats mean(petal*) as petal* by cluster 
    | eval species = "Centroid: ".cluster
    | eval point_size = 2] 
| fields species  petal* point_size

alt text

aoliner_splunk
Splunk Employee
Splunk Employee

You can inspect the KMeans model you built with fit using the summary command:

| summary <your_model_name>

Although, if you're trying to assign new points to the appropriate cluster, you can simply apply your model:

<new_points> | apply <your_model_name>

kiril123
Path Finder

This is great. Thank you.

0 Karma
Get Updates on the Splunk Community!

Devesh Logendran, Splunk, and the Singapore Cyber Conquest

At this year’s Splunk University, I had the privilege of chatting with Devesh Logendran, one of the winners in ...

There's No Place Like Chrome and the Splunk Platform

WATCH NOW!Malware. Risky Extensions. Data Exfiltration. End-users are increasingly reliant on browsers to ...

Customer Experience | Join the Customer Advisory Board!

Are you ready to take your Splunk journey to the next level? &#x1f680; We invite you to join our elite squad ...