文件列表:

rdp_nn.m

# rdp Recursive Deterministic Perceptron (RDP) to solve a non-linearly separable classification problem in MATLAB. A single-layer perceptron neural network is unable to solve non-linearly separable problems. However, the RDP can.
Training data - a set of nineteen 2element input vectors P and the corresponding nineteen 1element targets T - a non-linearly separable classification problem:

P = [ 2 1 2 5 7 2 3 6 1 2 5 4 6 5 7 8 7;
2 3 3 3 3 4 4 4 5 5 5 6 6 7 6 7 7 ];
T = [ 0 0 0 1 1 0 0 1 0 0 1 1 1 1 0 0 0 ];


As you can see visually, these data points are non-linearly separable.

The RDP works by first selecting a linearly separable subset from the original non-linearly separable problem. Then we create a perceptron to solve the linearly separable problem. Then we append the outputs/results of the trained perceptron to the input vectors of the original non-linearly separable problem, effectifvely adding a dimension (the input data changes from having 2 elements to 3 elements). The outputs of the first perceptron linearly separate the linearly separable subset from the rest of the data set. Lastly, we create a new perceptron with these new 3-element data inputs.


As you can see visually, these data points are now linearly separable with the added dimension.

Visualisation of the hyperplane that correctly separates the different classes:

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：