def do_bayes_opt_for(n):
svcBO = BayesOptCV(svccv, param_grid={'C':{'type':'float', 'min':min_C, 'max':max_C}},
bigger_is_better=True, verbose=2)
# svcBO.initialize(num_init=10, init_grid={'C': [1, 15, 60]})
svcBO.initialize(num_init=3)
kernel_param = {'theta0':0.5}
acqui_param = {'kappa':2}
svcBO.optimize(kernel_param=kernel_param, acqui_param=acqui_param, n_iter=n - 3, acqui_type='ucb', n_acqui_iter=200)
return svcBO.get_best()
scaler = MinMaxScaler(feature_range=(-1, 1))
train_data = scaler.fit_transform(train_data.toarray())
test_data = scaler.transform(test_data.toarray())
print train_data.shape
print test_data.shape
import numpy as np
f = lambda x: np.sin(x)
sigma = 0.02 # noise variance.
def sin(C):
return f(C) + sigma * np.random.randn(1)[0]
def svccv(C):
return 1 - cross_val_score(SVC(C=C, kernel='poly', degree=3 , random_state=2),
train_data, train_labels, 'f1', cv=5).mean()
svcBO = BayesOptCV(svccv, param_grid={'C':{'type':'float', 'min':6.4e-05, 'max':60}},
bigger_is_better=False, verbose=2)
svcBO.initialize(num_init=10, init_grid={'C': [1, 15, 60], 'gamma':[1, 5, 10]})
# kernel_param = {'nugget':0.0000001}
kernel_param = {'theta0':0.5}
acqui_param = {'kappa':4}
svcBO.optimize(kernel_param=kernel_param, acqui_param=acqui_param, n_iter=25, acqui_type='ucb', n_acqui_iter=200)
print('Final Results')
print('SVC: %f' % svcBO.report['best']['best_val'])
plt.show()
def svccv(C):
return 1 - cross_val_score(SVC(
C=C, kernel='poly', degree=3, random_state=2),
train_data,
train_labels,
'f1',
cv=5).mean()
svcBO = BayesOptCV(
svccv,
param_grid={'C': {
'type': 'float',
'min': 6.4e-05,
'max': 60
}},
bigger_is_better=False,
verbose=2)
svcBO.initialize(num_init=10,
init_grid={
'C': [1, 15, 60],
'gamma': [1, 5, 10]
})
# kernel_param = {'nugget':0.0000001}
kernel_param = {'theta0': 0.5}
acqui_param = {'kappa': 4}
svcBO.optimize(kernel_param=kernel_param,
acqui_param=acqui_param,
