def run_regression_1D_collapsed():
np.random.seed(42)
print "create dataset ..."
Xtrain, ytrain, Xtest, ytest = create_dataset()
alphas = [0.001, 0.1, 0.2, 0.3, 0.5, 0.7, 0.8, 1]
for alpha in alphas:
M = 20
model = vfe.SGPR_collapsed(Xtrain, ytrain, M)
model.optimise(method='L-BFGS-B', alpha=alpha, maxiter=1000, disp=False)
my, vy = model.predict_y(Xtest, alpha)
my = np.reshape(my, ytest.shape)
vy = np.reshape(vy, ytest.shape)
rmse = np.sqrt(np.mean((my - ytest)**2))
ll = np.mean(-0.5 * np.log(2 * np.pi * vy) - 0.5 * (ytest - my)**2 / vy)
nlml, _ = model.objective_function(model.get_hypers(), alpha)
print 'alpha=%.3f, train ml=%3f, test rmse=%.3f, ll=%.3f' % (alpha, nlml, rmse, ll)
# plot(model, Xtrain, ytrain)
# plt.show()
# should produce something like this
# alpha=0.001, train ml=-64.573021, test rmse=0.169, ll=0.348
# alpha=0.100, train ml=-64.616618, test rmse=0.169, ll=0.348
# alpha=0.200, train ml=-64.626655, test rmse=0.169, ll=0.348
# alpha=0.300, train ml=-64.644053, test rmse=0.169, ll=0.348
# alpha=0.500, train ml=-64.756588, test rmse=0.169, ll=0.348
# alpha=0.700, train ml=-68.755871, test rmse=0.169, ll=0.350
# alpha=0.800, train ml=-72.153441, test rmse=0.167, ll=0.349
# alpha=1.000, train ml=-71.305002, test rmse=0.169, ll=0.303
评论列表
文章目录
