New to AutoModel, trying to use Clustering
I see both k-means and x-means on my screen shot. How do I play around with k-Means and x-Means? After I click on run, how will I know if there is an optimal number for k?
Find more posts tagged with
Sort by:
1 - 2 of
21
Auto Model in general does not optimise k-Means, you need to specify the number of clusters (the screen you have attached shows the field to enter k). x-Means automatically selects the best k within a range, which in you case (default) is between 2 and 20. You can see what was the selected number of clusters by checking x-Means summary report. Before running clustering you may also request automatic feature selection, which is likely to improve clustering by removing highly correlated / linearly dependent attributes (it takes some time to do so on large data sets). In general, you'd be better off doing data pre-processing yourself, e.g. selecting independent attributes and normalising them before k-Means. Jacob
BTW, I did not realise that Auto Model builds a cluster tree, which I assume is simply a Decision Tree built on the selected (and engineered) attributes as predictors and using the clusters as labels. Very interesting new trend in RM adopting the principles of explainable machine learning! (considering that cluster models are black-boxes)
Sort by:
1 - 1 of
11
BTW, I did not realise that Auto Model builds a cluster tree, which I assume is simply a Decision Tree built on the selected (and engineered) attributes as predictors and using the clusters as labels. Very interesting new trend in RM adopting the principles of explainable machine learning! (considering that cluster models are black-boxes)