x = allx[:,
0::20, :] # picking every 20th term to get a total of 51 time points
t = allt[:, 0::20] # 51 time points, coarse data
data_param = prm.data(theta=0.5 * np.random.rand(prm.dof, prm.dim),
gvec=sim_param.gvec)
print("Data shape:", x.shape)
print("Theta shape:", data_param.theta.shape)
print("Theta:", data_param.theta)
# parvalue number of sub intervals
# Note : numsubinterval = 1 => only observed data points, no intermediate brownian bridges
em_param = prm.em(tol=0.01,
burninpaths=10,
mcmcpaths=100,
numsubintervals=parvalue,
niter=100,
dt=(allt[0, 1] - allt[0, 0]))
# call to EM which returns the final error and estimated theta value
error_list, theta_list, gammavec_list = em(x, t, em_param, data_param)
estimated_theta = prm.theta_transformations(theta=theta_list[-1],
theta_type='hermite')
true_theta = prm.theta_transformations(theta=dc.true_theta(sim_param),
theta_type='ordinary')
print("\n Estimated ordinary: ", np.transpose(estimated_theta.ordinary),
"\n True ordinary: ", np.transpose(true_theta.ordinary))
print("\n Estimated hermite: ", np.transpose(estimated_theta.hermite),
x = allx[:,
0::20, :] # picking every 20th term to get a total of 51 time points
t = allt[:, 0::20] # 51 time points, coarse data
data_param = prm.data(theta=0.5 * np.random.rand(prm.dof, prm.dim),
gvec=sim_param.gvec)
print("Data shape:", x.shape)
print("Theta shape:", data_param.theta.shape)
print("Theta:", data_param.theta)
# parvalue number of sub intervals
# Note : numsubinterval = 1 => only observed data points, no intermediate brownian bridges
em_param = prm.em(tol=0.01 * prm.dim * prm.dof,
burninpaths=10,
mcmcpaths=100,
numsubintervals=5,
niter=100,
dt=(allt[0, 1] - allt[0, 0]))
# call to EM which returns the final error and estimated theta value
error_list, theta_list, gammavec_list = em(x, t, em_param, data_param)
estimated_theta = prm.theta_transformations(theta=theta_list[-1],
theta_type='hermite')
true_theta = prm.theta_transformations(theta=dc.true_theta(sim_param),
theta_type='ordinary')
print("\n Estimated ordinary: ", np.transpose(estimated_theta.ordinary),
"\n True ordinary: ", np.transpose(true_theta.ordinary))
print("\n Estimated hermite: ", np.transpose(estimated_theta.hermite),
# load data
import pickle
with open('nem_6D.pkl', 'rb') as f:
allx, allt, x_without_noise, euler_param, sim_param = pickle.load(f)
data_param = prm.data(theta=0.5 * np.random.rand(prm.dof, prm.dim),
gvec=sim_param.gvec)
print("Data shape:", allx.shape)
print("Theta shape:", data_param.theta.shape)
print("Theta:", data_param.theta)
"""
Default parameters for Expectation-Maximization
em_param = param.em(tol = 1e-3, burninpaths = 100, mcmcpaths = 1000, numsubintervals = 10, niter = 100, dt = (allt[0, 1] - allt[0, 0]))
"""
em_param = prm.em(dt=allt[0, 1] - allt[0, 0])
# call to EM which returns the final error and estimated theta value
error_list, theta_list = nb.em(allx, allt, em_param, data_param)
# convert the hermite polynomial to simplified polynomial expression
# final_parameters = nb.hermite_to_polynomial(theta_list[-1])
print("Error: ", error_list)
print("Theta: ", theta_list)
# print("Final parameters: ", final_parameters)
# save to file
with open('6D.pkl', 'wb') as f:
pickle.dump(
[error_list, theta_list, final_parameters, em_param, data_param], f)
