def eval_GP(m,X_test, Y_test):
xx = X_test
mean, var = m.predict_y(xx)
percentage = 0.0
for ii in range(0,len(xx[:,0])):
if ((Y_test[ii] <= (mean[ii] + 2.0*np.sqrt(var[ii]))) and (Y_test[ii] >= (mean[ii] - 2.0*np.sqrt(var[ii])))):
percentage += 1.0
percentage /= len(xx[:,0])
v_ll = Gauss_log_likelihood(np.expand_dims(Y_test,axis=1),(mean - 2.0*np.sqrt(var)), (mean + 2.0*np.sqrt(var)))
ll = np.sum(v_ll)
v_SE = (Y_test - np.squeeze(mean))**2
RMSE = np.mean(v_SE)**(0.5)
return RMSE,ll, percentage
def eval_BNN(m,X_test, Y_test):
xx = X_test
M_pred = np.zeros((samples,len(xx[:,0])))
for ii in range(0,samples):
M_pred[ii,:] = np.squeeze(m.predict(xx))
y_mean = np.mean(M_pred, axis=0)
y_var = np.var(M_pred, axis=0)
percentage = 0.0
for ii in range(0,len(xx[:,0])):
if ((Y_test[ii] <= (y_mean[ii] + 2.0*np.sqrt(y_var[ii]))) and (Y_test[ii] >= (y_mean[ii] - 2.0*np.sqrt(y_var[ii])))):
percentage += 1.0
percentage /= len(xx[:,0])
v_ll = Gauss_log_likelihood(Y_test,(y_mean - 2.0*np.sqrt(y_var)), (y_mean + 2.0*np.sqrt(y_var)))
ll = np.sum(v_ll)
v_SE = (Y_test - y_mean)**2
RMSE = np.mean(v_SE)**(0.5)
return RMSE,ll, percentage