Browse through the JavaScript code to see how the data attributes are processed into a distance measure. Pay particular attention to normalizing and the way the attributes are combined into a single measurement of distance.
Save both the HTML and the JS file on your computer so you can edit the code. Within the distance calculations, add weights to the attributes so that the color has a higher impact than than the position and that the size has a very small impact on similarity.
With before and after screenshots or videos, document the effect of your modifications in the behavior of the clustering.
Investigate methods to express qualitative similarity of non-numerical data (text, images, sound, opinions, etc.) in numerical terms. You can take a look at an example textbook or search for sources on your own.
Report your findings in writing, with concrete examples of different kinds of non-numerical attributes and approaches to define numerical similarity measures for them. Remember to clearly cite all sources.
The value chosen for k affects the outcome of k-means clustering quite significantly. Investigate guidelines on how to choose it as well as automated methods for setting the value of k. Do not worry about the mathematical and computational details, but try to uncover the essence of the logic behind the choice.
Describe your findings in writing with your own words, citing all of your sources.