#######################################
# pyGPGO examples
# acqzoo: shows the behaviour of different
# acquisition functions on a GP surrogate
# for a sine-like function
#######################################
import numpy as np
import matplotlib.pyplot as plt
from pyGPGO.surrogates.GaussianProcess import GaussianProcess
from pyGPGO.acquisition import Acquisition
from pyGPGO.covfunc import squaredExponential
from pyGPGO.GPGO import GPGO
def plotGPGO(gpgo, param, index, new=True):
param_value = list(param.values())[0][1]
x_test = np.linspace(param_value[0], param_value[1], 1000).reshape((1000, 1))
y_hat, y_var = gpgo.GP.predict(x_test, return_std=True)
std = np.sqrt(y_var)
l, u = y_hat - 1.96 * std, y_hat + 1.96 * std
if new:
plt.figure()
plt.subplot(5, 1, 1)
plt.fill_between(x_test.flatten(), l, u, alpha=0.2)
plt.plot(x_test.flatten(), y_hat)
plt.subplot(5, 1, index)
a = np.array([-gpgo._acqWrapper(np.atleast_1d(x)) for x in x_test]).flatten()
plt.plot(x_test, a, color=colors[index - 2], label=acq_titles[index - 2])
gpgo._optimizeAcq(method='L-BFGS-B', n_start=1000)
plt.axvline(x=gpgo.best)
plt.legend(loc=0)
if __name__ == '__main__':
def f(x):
return (np.sin(x))
acq_1 = Acquisition(mode='ExpectedImprovement')
acq_2 = Acquisition(mode='ProbabilityImprovement')
acq_3 = Acquisition(mode='UCB', beta=0.5)
acq_4 = Acquisition(mode='UCB', beta=1.5)
acq_list = [acq_1, acq_2, acq_3, acq_4]
sexp = squaredExponential()
param = {'x': ('cont', [0, 2 * np.pi])}
new = True
colors = ['green', 'red', 'orange', 'black']
acq_titles = [r'Expected improvement', r'Probability of Improvement', r'GP-UCB $\beta = .5$',
r'GP-UCB $\beta = 1.5$']
for index, acq in enumerate(acq_list):
np.random.seed(200)
gp = GaussianProcess(sexp)
gpgo = GPGO(gp, acq, f, param)
gpgo._firstRun(n_eval=3)
plotGPGO(gpgo, param, index=index + 2, new=new)
new = False
plt.show()
