def main():
# load GP model from checkpoint
gp, dataset = parse_and_load(CHECKPOINT_PATH, DATASET, INF, COV, FLAGS)
xtrain, ytrain = dataset.xtrain, dataset.ytrain
xtest, ytest = dataset.xtest, dataset.ytest
strain, stest = dataset.strain, dataset.stest
pred_mean, pred_var = gp.predict(np.concatenate((xtest, stest), -1))
out_dims = dataset.output_dim
fig = plt.figure()
for i in range(out_dims):
ax = fig.add_subplot(out_dims, 1, i + 1, projection='3d')
ax.scatter(xtrain[:, IN_DIM_A], strain[:, SENS_DIM], ytrain[:, i], marker='.', s=40, label='trainings')
ax.scatter(xtest[:, IN_DIM_A], stest[:, SENS_DIM], ytest[:, i], marker='o', s=40, label='tests')
ax.scatter(xtest[:, IN_DIM_A], stest[:, SENS_DIM], pred_mean[:, i], marker='x', s=40, label='predictions')
upper_bound = pred_mean[:, i] + 1.96 * np.sqrt(pred_var[:, i])
lower_bound = pred_mean[:, i] - 1.96 * np.sqrt(pred_var[:, i])
# ax.add_collection3d(plt.fill_between(xtest[:, IN_DIM_A], lower_bound, upper_bound, color='gray', alpha=0.25,
# label='95% CI'), zs=stest[:, SENS_DIM], zdir='y')
# legend doesn't work with `add_collection3d`
# plt.legend(loc='lower left')
if SAVE_ANIMATION:
def _rotate(angle):
ax.view_init(azim=angle)
rot_animation = animation.FuncAnimation(fig, _rotate, frames=np.arange(0, 362, 2), interval=100)
rot_animation.save('rotation.gif', dpi=80, writer='imagemagick')
else:
plt.show()
def main():
# load GP model from checkpoint
gp, _ = parse_and_load(CHECKPOINT_PATH, DATASET, INF, COV, FLAGS)
# get variables
var_collection = {var.name: var.numpy() for var in gp.variables}
# save in file
np.savez_compressed(SAVE_PATH, **var_collection)
def main():
# load GP model from checkpoint
gp, _ = parse_and_load(CHECKPOINT_PATH, DATASET, INF, COV, FLAGS)
# get variables
hyper_params = gp.lik.get_params() + sum([k.get_params()
for k in gp.cov], [])
var_collection = {
var.name: var.numpy()
for var in gp.get_all_variables() + hyper_params
}
# save in file
np.savez_compressed(SAVE_PATH, **var_collection)
def main():
# load GP model from checkpoint
gp, dataset = parse_and_load(CHECKPOINT_PATH, DATASET, INF, COV, FLAGS)
# make predictions
pred_mean, pred_var = gp.predict({'input': dataset.xtest})
# save in file
if AS_TXT_FILE:
full_path = RESULT_PATH + ".txt"
np.savetxt(full_path,
np.column_stack((dataset.xtest[:, 0], pred_mean.numpy()[:,
0],
pred_var.numpy()[:, 0])),
header='xtest, pred_mean, pred_var')
else:
np.savez_compressed(RESULT_PATH,
pred_mean=pred_mean,
pred_var=pred_var)