$k$-Means Clustering

$k$-Means Clustering

Normally we would not (could not) compute accuracy on an unsupervised algorithm

like $k$-means clustering because there is no $y$.

However, I chose to cluster the digits images, for which we have $y$ labels. Here I check whether the clusters correspond to the $y$ labels I know.

Notice that the clusters are such that for each digit $i$ there is a cluster dominated by that digit. Some details:

Again, remember that clustering is normally done on unlabeled data, not on these labeled images of digits; and remember that I cheated to choose $k=10$.