import numpy as np
import matplotlib.pyplot as plt
import rdml_graph as gr
if __name__ == '__main__':
X_train = np.array([0,1,2,3,4.2,6,7])
ratings = np.array([5,5,2,1,2 ,3,3])
gp = gr.PreferenceGP(gr.RBF_kern(0.5, 0.7), \
other_probits={'ordinal': gr.OrdinalProbit(2.0,1.0, n_ordinals=5)})
#gp = gr.PreferenceGP(gr.periodic_kern(1.2,0.3,5))
#gp = gr.PreferenceGP(gr.linear_kern(0.2, 0.2, 0.2))
#gp = gr.PreferenceGP(gr.RBF_kern(0.2,1)+gr.periodic_kern(1,0.2,0)+gr.linear_kern(0.2,0.1,0.3))
#gp = gr.PreferenceGP(gr.RBF_kern(0.1,1)*gr.linear_kern(0.3,0.2,0.3))
gp.add(X_train, ratings, type='ordinal')
#gp.optimize(optimize_hyperparameter=True)
#print('gp.calc_ll()')
#print(gp.calc_ll())
X = np.arange(-0.5, 8, 0.1)
mu, sigma = gp.predict(X)
import rdml_graph as gr
if __name__ == '__main__':
X_train = np.array([0,1,2,3,6,7])
X = np.arange(-3, 12, 0.1)
y_train = np.array([1, 0.5,0, -1, 1, 2])
#y_train = np.array([0.2, 0.3, 0.4, 0.52, 0.7, 0.76])
#training_sigma = 0
training_sigma=np.array([1, 0.5, 0.1, 0.1, 0.2, 0])
#gp = gr.GP(gr.RBF_kern, {'rbf_sigma': 1, 'rbf_l': 1})
#gp = gr.GP(gr.periodic_kern, {'periodic_sigma': 1, 'periodic_l': 1, 'periodic_p': 20})
#gp = gr.GP(gr.linear_kern, {'linear_sigma': 5, 'linear_sigma_b': 5, 'linear_offset': 0.2})
gp = gr.GP(gr.RBF_kern(1,1)+gr.periodic_kern(1,1,10)+gr.linear_kern(3,1,0.3))
gp.add(X_train, y_train, training_sigma=training_sigma)
mu, sigma = gp.predict(X)
std = np.sqrt(sigma)
plt.plot(X, mu)
sigma_to_plot = 1
plt.gca().fill_between(X, mu-(sigma_to_plot*std), mu+(sigma_to_plot*std), color='#dddddd')
plt.scatter(X_train, y_train)
plt.title('Gaussian Process estimate (1 sigma)')
plt.xlabel('x')
plt.ylabel('y')
plt.show()
# TestGPKernels.py # Written Ian Rankin October 2021 # # import numpy as np import rdml_graph as gr import pdb rbf = gr.RBF_kern(1, 1) kern2 = gr.periodic_kern(1, 1, 3) kern3 = gr.linear_kern(1, 1, 1) print(rbf) print(kern2) print(kern3) print(rbf(1, 2)) print(kern2(1, 2)) print(kern3(1, 2)) combined = rbf + (kern2 * kern3) print(combined) print(combined(1, 2)) print(combined.get_param()) print(combined.gradient(1, 2)) combined.set_param([2, 3, 5, 4, 7, 1, 1, 2]) print(combined.get_param()) x = np.array([0, 1, 4, 5, 6, 7])
# Written Ian Rankin - September 2021
#
# An example of using the Preference GP with the absbounded probit.
import numpy as np
import matplotlib.pyplot as plt
import rdml_graph as gr
if __name__ == '__main__':
X_train = np.array([0, 1, 2, 3, 4.2, 6, 7])
abs_values = np.array([0.8, 0.6, 0.3, 0.2, 0.22, 0.4, 0.5])
#abs_values = np.array([0.4, 0.2, 0.2, 0.2, 0.1, 0.11, 0.3])
gp = gr.PreferenceGP(gr.RBF_kern(0.5, 0.7), normalize_gp=False, \
normalize_positive=False, \
other_probits={'abs': gr.AbsBoundProbit(1.0,5.0)})
#gp = gr.PreferenceGP(gr.periodic_kern(1.2,0.3,5))
#gp = gr.PreferenceGP(gr.linear_kern(0.2, 0.2, 0.2))
#gp = gr.PreferenceGP(gr.RBF_kern(0.2,1)+gr.periodic_kern(1,0.2,0)+gr.linear_kern(0.2,0.1,0.3))
#gp = gr.PreferenceGP(gr.RBF_kern(0.1,1)*gr.linear_kern(0.3,0.2,0.3))
gp.add(X_train, abs_values, type='abs')
#gp.optimize(optimize_hyperparameter=True)
#print('gp.calc_ll()')
#print(gp.calc_ll())
X = np.arange(-0.5, 8, 0.1)
mu, sigma = gp.predict(X)
if __name__ == '__main__':
num_side = 25
bounds = [(0,7), (0,7)]
num_train_pts = 40
num_alts = 4
utility_f = f_sq
#gp = gr.PreferenceGP(gr.RBF_kern(0.2,0.5)*gr.linear_kern(0.2, 0.1, 0))
#gp = gr.PreferenceGP(gr.linear_kern(0.3, 0.1, 0.0))
gp = gr.PreferenceGP(gr.RBF_kern(1.0, 1.0), pareto_pairs=True, \
use_hyper_optimization=False, \
active_learner = gr.DetLearner(0.9))
gp.add_prior(bounds=np.array(bounds), num_pts=20)
for i in tqdm.tqdm(range(10)):
train_X = np.random.random((num_train_pts,2)) * np.array([bounds[0][1]-bounds[0][0], bounds[1][1]-bounds[1][0]]) + np.array([bounds[0][0], bounds[1][0]])
train_Y = utility_f(train_X)#f_lin(train_X)
selected_idx, UCB, best_value = gp.select(train_X, num_alts)
best_idx = np.argmax(train_Y[selected_idx])
pairs = gr.ranked_pairs_from_fake(train_X[selected_idx], utility_f)
def f_lin(x, data=None):
return x[:, 0] * x[:, 1]
if __name__ == '__main__':
num_side = 25
bounds = [(0, 7), (0, 7)]
num_train_pts = 50
train_X = np.random.random((num_train_pts, 2)) * np.array(
[bounds[0][1] - bounds[0][0], bounds[1][1] - bounds[1][0]]) + np.array(
[bounds[0][0], bounds[1][0]])
train_Y = f_sin(train_X)
gp = gr.GP(gr.RBF_kern(1, 0.8))
for i in tqdm.tqdm(range(20)):
train_X = np.random.random((num_train_pts, 2)) * np.array([
bounds[0][1] - bounds[0][0], bounds[1][1] - bounds[1][0]
]) + np.array([bounds[0][0], bounds[1][0]])
train_Y = f_sin(train_X)
selected_idx, UCB, best_value = gp.ucb_selection(train_X, 5)
#print(selected_idx)
#print(UCB)
#print(train_X[selected_idx])
#pdb.set_trace()
gp.add(train_X[selected_idx], train_Y[selected_idx])
#gp.add(train_X, train_Y)