Interactive k-means

Educational widgets that shows working of a k-means clustering.

Signals

Inputs

• Data

Input data set.

Outputs

• Data

Data set with clusters labels annotation.

• Centroids

Centroids position

Description

The aim of this widget is to show the working of a k-means clustering algorithm on two attributes from data set. Widget applies k-means clustering to the selected two attributes step by step. User can step through the algorithm and see how the algorithm works.

1. Select attribute for x axis and attribute for y axis.
2. Select number of centroids in the spinner. If you want new random positions of the centroids or restart the algorithm in any point, you can click on Randomize button. If you want to add centroid on particular position in the graph just click on this position. If you want to move centroid, grab it and drop it on the desired position.
3. Step through the algorithm with Recompute centroids and Reassign membership. If you want to make step back use Step back button. You can also step automatically with pressing on Run button. Speed spinner can be used to set the speed of automatic stepping.

Example

Here are two possible schemas that shows how the Interactive k-Menas widget can be used. You can load data form File or use any other data source, such as Paint Data. Interactive k-Menas widget also produce the data table with results of clustering and table with centroids positions. That can be inspected with Data Table widget or andy other widgets (e.g. Scatter Plot).

Let us demonstrate the working of the widget on Iris Dataset.

We provide data from data set using File widget. Then we open Interactive k-Means widget. We decided that we will demonstrate k-Menas on petal length and petal width attributes, so we set them as X and Y parameters. We also decided to perform a clustering in 3 clusters. This is set as Number of centroids. After that we get this figure.

If we are not satisfied with positions of centroids we can change it with click on Randomize button. Then we perform first recomputing of centroids with click on Recompute centroids button. We get following image.

The next step is to reassign membership of point to the closest centroid. It is performed with click on Reasign membership button.

Then we repeat that two steps until algorithm converges. It is final result.

But we are not satisfied with result because we noticed that maybe classification in 4 clusters would be better. So we decided to add a new centroid. We can make it with increasing number of centroids in left menu or with click on position in graph where we want to place the centroid. We decided to add it with click. New centroid is red one.

Now we can repeat some steps that algorithm converges again, but before we will move green centroid to change the behavior of the algorithm. We grab green centroid and moved it to desired position.