def meta_val(x_pred, y_real, x_exp, data, x_prob, a, setting, lower_bound,
upper_bound):
if len(a) == 0:
setting['exp_x'] = x_exp
setting['exp_y'] = data[:, 2]
setting['bounds'] = [(lower_bound, upper_bound),
(lower_bound, upper_bound),
(lower_bound, upper_bound)]
model = createKriging.createKriging(setting)
y_kriging, ci_kriging, var_kriging = createKriging.predictor(
x_pred, model, setting)
y_pred = y_kriging
y_Real = y_real[:, 2]
rmse, nrmse = utilPCE.validation(y_Real, y_pred)
else:
rmse = npy.zeros(len(a))
nrmse = npy.zeros(len(a))
for i in xrange(0, len(a)):
setting['exp_x'] = x_exp
setting['exp_y'] = data[i * len(x_exp):(i + 1) * len(x_exp), 2]
setting['bounds'] = [(lower_bound, upper_bound),
(lower_bound, upper_bound)]
model = createKriging.createKriging(setting)
y_kriging, ci_kriging, var_kriging = createKriging.predictor(
x_pred, model, setting)
y_pred = y_kriging
y_Real = y_real[i * len(x_pred):(i + 1) * len(x_pred), 2]
rmse[i], nrmse[i] = utilPCE.validation(y_Real, y_pred)
return rmse, nrmse
def meta_gen(x_pred, x_exp, data, x_prob, a, setting, lower_bound, upper_bound):
if len(a) == 0:
setting['exp_x'] = x_exp
setting['exp_y'] = data[:, 2]
setting['bounds'] = [(lower_bound, upper_bound), (lower_bound, upper_bound), (lower_bound, upper_bound)]
model = createKriging.createKriging(setting)
y_kriging, ci_kriging, var_kriging = createKriging.predictor(x_pred, model, setting)
y_pred = y_kriging.reshape(len(x_pred), 1)
y_pred[y_pred < 0.0001] = min(y_pred[y_pred > 0.0001])
index = npy.arange(1, len(x_pred) + 1)
index.shape = (len(x_pred), 1)
x_a = x_pred[:, 0].reshape(len(x_pred), 1)
data_meta = npy.concatenate((index, x_a, y_pred), axis=1)
else:
data_meta = npy.zeros((len(a) * len(x_pred), 3))
for i in xrange(0, len(a)):
setting['exp_x'] = x_exp
setting['exp_y'] = data[i * len(x_exp):(i + 1) * len(x_exp), 2]
setting['bounds'] = [(lower_bound, upper_bound), (lower_bound, upper_bound)]
model = createKriging.createKriging(setting)
y_kriging, ci_kriging, var_kriging = createKriging.predictor(x_pred, model, setting)
y_pred = y_kriging.reshape(len(x_pred), 1)
y_pred[y_pred < 0.0001] = min(y_pred[y_pred > 0.0001])
index = npy.arange(1, len(x_pred) + 1)
index.shape = (len(x_pred), 1)
data_meta[i * len(x_pred):(i + 1) * len(x_pred), :] = npy.concatenate(
(index, a[i] * npy.ones((len(x_pred), 1)), y_pred), axis=1)
data = data_meta[data_meta[:, 2] > min(data_meta[:, 2])]
return data
def sobol_gen(a, x_prob, dim, n_exp, x_mix, *args):
x_exp1 = args[0]
x_exp2 = args[1]
x_exp = args[2]
data = args[3]
setting = args[4]
upper_bound = args[5]
lower_bound = args[6]
y = npy.zeros((n_exp, dim))
y1 = npy.zeros((n_exp, 1))
y2 = npy.zeros((n_exp, 1))
setting['exp_x'] = x_exp
setting['exp_y'] = data[:, 2]
if len(a) == 0:
setting['bounds'] = [(lower_bound, upper_bound),
(lower_bound, upper_bound),
(lower_bound, upper_bound)]
else:
setting['bounds'] = [(lower_bound, upper_bound),
(lower_bound, upper_bound)]
model = createKriging.createKriging(setting)
y1, ci_kriging, var_kriging = createKriging.predictor(
x_exp1, model, setting)
y2, ci_kriging, var_kriging = createKriging.predictor(
x_exp2, model, setting)
for i in xrange(0, dim):
x_mix[:, :] = x_exp1
x_mix[:, i] = x_exp2[:, i]
y[:, i], ci_kriging, var_kriging = createKriging.predictor(
x_mix, model, setting)
return y1, y2, y
def sobol_gen(a, x_prob, dim, n_exp, x_mix, *args):
x_exp1 = args[0]
x_exp2 = args[1]
n_deg = args[2]
x_exp = args[3]
data = args[4]
setting = args[5]
upper_bound = args[6]
lower_bound = args[7]
y = npy.zeros((n_exp, dim))
y1 = npy.zeros((n_exp, 1))
y2 = npy.zeros((n_exp, 1))
setting['exp_x'] = x_exp
setting['exp_y'] = data[:,2]
y_exp = data[:,2]
PCE_lars = calPCE.collocation(n_deg, x_prob, x_exp, y_exp, 'LARS')
if len(a) == 0:
setting['bounds'] = [(lower_bound, upper_bound), (lower_bound, upper_bound), (lower_bound, upper_bound)]
else:
setting['bounds'] = [(lower_bound, upper_bound), (lower_bound, upper_bound)]
model = createKriging.createKriging(setting, PCE_lars)
PCE_pred_1, y_PCE_pred = utilPCE.predictor(PCE_lars, x_exp1)
y1, ci_kriging, var_kriging = createKriging.predictor(x_exp1, model, setting, PCE_lars, PCE_pred_1)
PCE_pred_2, y_PCE_pred = utilPCE.predictor(PCE_lars, x_exp2)
y2, ci_kriging, var_kriging = createKriging.predictor(x_exp2, model, setting, PCE_lars, PCE_pred_2)
for i in xrange(0, dim):
x_mix[:,:] = x_exp1
x_mix[:,i] = x_exp2[:,i]
PCE_pred, y_PCE_pred = utilPCE.predictor(PCE_lars, x_mix)
y[:,i], ci_kriging, var_kriging = createKriging.predictor(x_mix, model, setting, PCE_lars, PCE_pred)
return y1, y2, y
