Advance Machine Learning Lab

Austin Schwinn and Jeremie Blanchard

M2 MLDM

Lab 1

Kernel Methods

For this lab, we compared kernelized and non-kernilzed forms on the following algorithms on non-linearly seperable datasets:

1. PCA
2. Kmeans
3. Logisitic Regression

Using the following kernel tricks:

1. Linear
2. Gaussian RBF
3. Polynomial
4. Laplacian

We compared these algorithms using the following benchmark datasets from scikit learn:

1. Half Moons
2. Center Circles
3. Swiss Rolls
4. Classification

For each of the listed algorithms and datasets, we compared our own implimentation against premade scikit-learn's implimentations. We compared the quality of the results, such as the explained variance with PCA or the accuracy of logisitic regression, as well as the run-time efficiency.

In part 4 of the lab, we impliment one class SVM and Maximum Enclosing Ball in MatLab.