The results are not as good as the ones we got in the previous recipe. The group of outliers on the lower-left part of the plot are now assigned to a cluster, instead of being flagged as outliers. Moreover, tuning the q parameter seems quite challenging.

Nevertheless, the covariance matrix is adjusted according to the q parameter, which means that the estimated densities differ in the two cases. The smaller the q parameter is, the smaller the impact of each (outlier) observation on the covariance estimation. As we can see, we get densities that are obviously more concentrated as this q parameter decreases. 

Regardless of the q parameter, the nice aspect about this model is that it also provides an estimate of uncertainty. As we would expect, observations lying between classes are the ones with the highest uncertainty (the model is not very sure whether they should be in one class or another).