jpneto / markdowns Goto Github PK

This is one of the best tutorial repos for any language anywhere, makes it better for all of us who love R that it happens to be in the language we love the most. Please keep adding more if you can and thank you for sharing these with the world they are all truly fantastic.

In your first example, did you intend to use

exp(-gamma * norm(as.matrix(x),"F") )

or

exp(-gamma * norm(as.matrix(x),"F")^2 )

i.e. the squared norm instead of just a simple euclidean distance?

I am asking this because it looks odd as you use the squared norm everywhere else and I don't see any special emphasis on a simple euclidean distance in your document (also here http://www.di.fc.ul.pt/~jpn/r/rbf/rbf.html).

Thank you very much in advance.

Your page here:
http://www.di.fc.ul.pt/~jpn/r/fourier/fourier.html
Is not rendering LaTeX code, rather it is displaying raw LaTeX code.
I downloaded the Rmd and csv files and managed to "knitHTML" and "knitPDF" with LaTeX displaying correctly. The only fault in my attempt was your accented name.
Thank you for a very useful tutorial (I have sent a link to my son who is studying physics at school and writing an assignment on guitar string tension and resulting tones.

Hi jpneto,

Your implementation of the Radial Basis Function net was performed using a uniformly distributed training and test datasets. As you are already aware, the main role of the basis function is essentially mixture density estimation of the input space, hence with a K = 10 centres, this will provide a good generalisation for this type of data configuration. However, using your implementation on a benchmark dataset like the Iris or MNIST in which the dataset configuration is not so uniformly defined, one will potentially obtain a very poor generalisation following such ad hoc implementation of the basis function. So I do think that a more principled network training approach for the basis function like the one in page 170 of bishop's book will provide reliability for generalisation: http://cs.du.edu/~mitchell/mario_books/Neural_Networks_for_Pattern_Recognition_-_Christopher_Bishop.pdf

Thank you

Kingsley.