What does the output of an a Neural Network with soft-max and RBF kernels look like?

rfb_nn.py

import numpy as np
from scipy.stats import norm
import pylab as pl


def rbf(x, C, w):
  y = [norm.pdf(np.linalg.norm(x - C[i]) ** 2.0, 0, 1) for i in range(C.shape[0])]
  y = np.exp(y) / np.sum(np.exp(y))
  return w.dot(y)


if __name__ == '__main__':
  C = np.random.randn(2, 2)
  w = np.random.randn(2)

  n = 200
  xs = ys = np.linspace(-2, 2, n)
  xx, yy = np.meshgrid(xs, ys)
  zz = np.zeros(xx.shape)
  for i in range(n):
    for j in range(n):
      p = [xx[i,j], yy[i,j]]
      zz[i,j] = rbf(p, C, w)

  ax = Axes3D(pl.gcf())
  ax.plot_surface(xx, yy, zz)