The best characterization I have heard recently distinguished between
"traditional statistics" and "data mining".  The key factor
in the distinction was that in traditional statistics, you test hypotheses
against data whereas in data mining you generate hypotheses (called models)
from the data.

In my view, machine learning is pretty closely synonymous with data mining
and the key distinction is learning from the data.  If you exclude LDA, then
you exclude k-means (which is essentially the same algorithm), but both are
classic unsupervised learning applications.  FPgrowth is in much the same
category as clustering.

I think it is a mistake to assume that only supervised learning is machine
learning.

On Tue, May 11, 2010 at 2:08 AM, Robin Anil <[email protected]> wrote:

> Just a thought. "scalable machine learning and data-mining libraries" ?.
> FPgrowth is not machine learning. Similary LDA is not machine learning but
> more like data modelling. I know, its all fuzzy, and wish we had a better
> way to say it. "tools for understanding patterns from data and predicting
> from learned ones"
>

Reply via email to