Z = np.column_stack((np.ones(nobs_i), Z))
noise = 0.1 * np.random.randn(nobs_i) #sig_e = 0.1
#generate endogenous variable
Y = np.dot(X, beta) + np.dot(Z, gamma_re) + noise
#add random effect design matrix also to fixed effects to
#capture the mean
#this seems to be necessary to force mean of RE to zero !?
#(It's not required for estimation but interpretation of random
#effects covariance matrix changes - still need to check details.
X = np.hstack((X, Z))
#create units and append to list
unit = Unit(Y, X, Z)
units.append(unit)
m = OneWayMixed(units)
import time
t0 = time.time()
m.initialize()
res = m.fit(maxiter=100, rtol=1.0e-5, params_rtol=1e-6, params_atol=1e-6)
t1 = time.time()
print 'time for initialize and fit', t1 - t0
print 'number of iterations', m.iterations
#print dir(m)
#print vars(m)
print '\nestimates for fixed effects'
print m.a
noise = 0.1 * np.random.randn(nobs_i) #sig_e = 0.1
#generate endogenous variable
Y = np.dot(X, beta) + np.dot(Z, gamma_re) + noise
#add random effect design matrix also to fixed effects to
#capture the mean
#this seems to be necessary to force mean of RE to zero !?
#(It's not required for estimation but interpretation of random
#effects covariance matrix changes - still need to check details.
#X = np.hstack((X,Z))
X = np.hstack((X, time_dummies))
# create units and append to list
new_unit = Unit(Y, X, Z)
units.append(new_unit)
m = OneWayMixed(units)
import time
t0 = time.time()
m.initialize()
res = m.fit(maxiter=100, rtol=1.0e-5, params_rtol=1e-6, params_atol=1e-6)
t1 = time.time()
print('time for initialize and fit', t1 - t0)
print('number of iterations', m.iterations)
#print dir(m)
#print vars(m)
print('\nestimates for fixed effects')
print(m.a)