Secondary Protein Prediction Using a Simple Perceptron

[S_7]

I am trying to understand how protein secondary prediction is working.
I know that each time you choose a number of window size to work with.
Lets say i choose number 5.
so i will have from my whole protein
...GHGVV... right?
i encode each of them as in 1-20 binary vectors.
that is, each amino acid is represented by a vector.
G = [ 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0]... for example.

now the thing that i don't understand, is how alpha-helices are classified, using a simple perceptron.
how can I know that G, or H, or V is an alpha-helix just by classifying the protein into 1-20 vectors?

thx for the help!