def plot_noise(figure_id_start, X, noise, title):
plt.figure(figure_id_start)
n_plots = noise.shape[1] // 2
plt.suptitle(title)
for idx in range(n_plots):
plt.subplot(1, 3, idx + 1)
plt.scatter(slice_column(noise, 2 * idx),
slice_column(noise, 2 * idx + 1),
s=1,
c=X,
cmap='magma')
def plot_mse_values(model_means, igp_means, Y_true,
figure_id_start=0, initial_labels=None, plot_shape=(1, 3)):
labels = initialize_labels(Y_true.shape[1], initial_labels)
for out_id, label in enumerate(labels):
specify_plot_location(out_id, figure_id_start, plot_shape)
single_gpar_means = slice_column(model_means, out_id)
single_igp_means = slice_column(igp_means, out_id)
true_means = slice_column(Y_true, out_id)
gpar_smse = smse(true_means, single_gpar_means)
igp_smse = smse(true_means, single_igp_means)
plot_bar_plot([gpar_smse, igp_smse], ['GPAR', 'IGP'])
plt.title('{} SMSE'.format(label))
def plot_single_output(X, stacked_means, stacked_vars, out_id, label, display_var=False):
"""Construct plot containing the predictions and observations."""
means = slice_column(stacked_means, out_id)
plt.plot(X, means, label=label)
if display_var:
variances = slice_column(stacked_vars, out_id)
ub = means + 2 * np.sqrt(variances)
lb = means - 2 * np.sqrt(variances)
plt.fill_between(
X.flatten(),
lb.flatten(),
ub.flatten(),
alpha=0.2,
edgecolor='b')
def _get_trained_gp_model(self, current_X, out_id):
y = slice_column(self.Y_obs, out_id)
kernel = self._get_kernel(self.X_obs, current_X)
m = self._get_model(current_X, y, kernel)
m.likelihood.variance = self.init_likelihood_var
if self.is_zero_noise:
m.likelihood.variance = 0.00001
m.likelihood.variance.trainable = False
self._optimize_model(m)
return m
def augment_X(self, current_X, out_id):
if current_X is None:
return self.X_obs
y = slice_column(self.Y_obs, out_id)
return concat_right_column(current_X, y)
def plot_truth(X_new, Y_true, out_id):
single_Y = slice_column(Y_true, out_id)
plt.plot(X_new, single_Y, label='Truth')
def plot_observations(X_obs, Y_obs, out_id):
single_Y = slice_column(Y_obs, out_id)
plt.scatter(X_obs, single_Y, color='b', marker='x', label='Observations')