nn = NeuralNetwork(sizes=[1, 5, 1], activation_function="sigmoid")
y = nn.eval(1.0)
print y
gamma = np.random.randn(nn.n) * 0.2
for j in range(10000):
nn.set_from_array(gamma)
dJdgamma = np.zeros_like(gamma)
J = 0.0
for i in range(len(xd)):
xin = xd[i]
yeval = nn.eval(xin)
#print yeval
J += (yeval - yd[i])**2
dydgamma = nn.dydgamma(xin, gamma)
dJdgamma += 2.0 * (yeval - yd[i]) * dydgamma
gamma = gamma - dJdgamma / np.abs(dJdgamma) * 0.001
if j % 100 == 0:
print j, J
#print gamma
yeval = nn.veval(xd)
figure()
plot(xd, yd, 'b.')
plot(xd, yeval, 'rx')
nn.save()
nn.load()
yeval = nn.veval(xd)
