today I want to document perceptron learning algorithm for classifications
Below goes following steps for teaching perceptron.
- Add one more column with value to 1 to each input row.
- Pick training cases according to some rules that gives you guarantee that every training case will be picked
- if output is correct, leave weights unchanged
- if ouput is 0 but 1 is expected then add input vector to weights of perceptron
- if output is 1 but 0 is expected then subtract input vector from weights of perceptron
This simple algorithm will find you a set of weights for right classification of your vector space if such set of weights exists. It will depend only from correctness of features that you use.