def generate_argmin_posterior(post_mean, post_covariance, sample_size):
sample = np.empty((sample_size, D))
unscaled_grid = unscale(grid, GP_model.FP.X_nonscaled)
for i in range(0, sample_size):
f_values = list(
np.random.multivariate_normal(post_mean,
post_covariance)) #predict/sample GP
argmax = unscaled_grid[np.where(
f_values == np.max(f_values))[0], :] #or unsclae grid before?
sample[i] = list(argmax[0])
return (sample)
grid = np.vstack([grid1.ravel(), grid2.ravel()])
else:
grid = np.vstack([grid, grid2.ravel()])
return (grid.T)
grid = create_grid(grid_x1, grid_x2, grid_x3, grid_x4, grid_x5, grid_x6)
del grid_x1, grid_x2, grid_x3, grid_x4, grid_x5, grid_x6
''' Slices plots of the posterior mean'''
slice_1_dim = 4
slice_2_dim = 5
fig = plt.figure()
ax = plt.axes(projection="3d")
ax.view_init(elev=90, azim=-90)
f_postmean = GP_model.mu_Sigma_pred(grid)[0]
unscaled_grid = unscale(grid, GP_model.FP.X_nonscaled)
ax.scatter3D(unscaled_grid[:, slice_1_dim - 1],
unscaled_grid[:, slice_2_dim - 1],
f_postmean,
c=f_postmean,
cmap='hsv')
ax.set_xlabel("\u03B1")
ax.set_ylabel("\u03B2")
plt.title('\u03BC($\mathbf{x}$)')
plt.show()
print(unscaled_grid[:, [slice_1_dim - 1, slice_2_dim -
1]][np.where(f_postmean == np.max(f_postmean))[0], :])
''' ------------------------------------------'''
''' Generate Argmin-distribution '''