Looking into the guts of things often reveals a very important perspective. Let me elaborate what I mean, via this anecdote:
Many machine learning practitioners will make a mistake of the following form:
"I ran a K-means clustering algorithm on my data, for k = 2, and it didnt show me anything interesting. Therefore, I conclude there isn't a good 2-clustering of my data."
There's a big problem here: if these practitioners knew what k-means does, they would realize the conclusion was off. K-means only works when the Euclidean metric is meaningful on their data. But, said practitioners have never looked into the guts of this method.
This example is from a 2nd year PhD student... (read 712 more words →)