Using Principal Component Analysis to find things that matter

Kernel PCA, in contrast to the PCA method that we just introduced, uses a user-defined kernel function to map the dataset with n dimensions to an m-dimensional feature space. PCA uses a linear function for the mapping and is equivalent to Kernel PCA with a linear kernel.

Kernel PCA can be especially useful if the data cannot be linearly separable so various nonlinear kernels can be used to map your data to higher dimensions.

Getting ready

To execute this recipe, you will need pandas and Scikit. No other prerequisites are required.

How to do it…

Once again, we wrap our model in a method so that we can track how long it takes for the model to converge. With Kernel PCA, you should expect significantly longer estimation times (the reduce_kernelPCA.py file):

@hlp.timeit
def reduce_KernelPCA(x, **kwd_params):
    '''
        Reduce the dimensions using Principal Component
        Analysis with different kernels
    '''
    # create the PCA...