# Perceptron Learning Algorithm

Is it possible to draw a hyperplane/decision boundary (not a decision function) from the parameters learned in PLA. If so how is this done (code would be ideal, if available). If not why?

The PLA returns the 'weights' vector orthogonal to the hyperplane you speak of. So, it is in a space of d-dimensions, where d is the number of features each data point has, and usually the 0th component denotes the bias or offset of this best fit hyperplane from the origin. You may need to normalize the final weights vector so that the (1,...,d) components are a unit (direction) vector, and then the 0th component is the offset.

• Very true, I have now read a lot of material in this since I asked the question. I am now stuck at the question that how learning would be affected by the fact that higher dimensional data actually are a union of disjoint lower manifolds of data. Jan 10 '19 at 18:54