def _checkdata(X,Y1_level,Y0_level,numAgents):
# get numCovarsOut from the ini.file.
initDict = grmReader.read()
Y1_beta_=initDict['Y1_beta']
numCovarsOut = np.array(Y1_beta_).shape[0]
# check X
assert(isinstance(X,np.ndarray))
assert(X.shape==(numAgents, numCovarsOut))
assert(np.all(np.isfinite(X)))
# check Y1_level
assert(isinstance(Y1_level,np.ndarray))
assert(Y1_level.shape==(numAgents, ))
assert(np.all(np.isfinite(Y1_level)))
# check Y0_level
assert(isinstance(Y0_level,np.ndarray))
assert(Y0_level.shape==(numAgents, ))
assert(np.all(np.isfinite(Y0_level)))
return True
def _getdata():
'''read the .dat file and export the simulated data
'''
# Process initialization file.
initDict = grmReader.read()
#read the data from the .dat file
data_ = np.genfromtxt(initDict['fileName'], dtype = 'float')
return data_
def simulate():
''' Simulate data generation process of the Generalized Roy Model. '''
print "Simulating Data."
# Process initFile.
initDict = grmReader.read()
#
# Distribute parametrization and (limited) type conversions.
#
numAgents = initDict['numAgents']
fileName = initDict['fileName']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
V_var = initDict['V_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
randomSeed = initDict['randomSeed']
# Set random seed
np.random.seed(randomSeed)
# Construct auxiliary objects.
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
U1V_cov = U1V_rho*np.sqrt(U1_var)*np.sqrt(V_var)
U0V_cov = U0V_rho*np.sqrt(U0_var)*np.sqrt(V_var)
#
# Simulation
#
# Simulate observable agent characteristics.
means = np.tile(0.0, numCovarsOut)
covs = np.identity(numCovarsOut)
X = np.random.multivariate_normal(means, covs, numAgents)
X[:,0] = 1.0
means = np.tile(0.0, numCovarsCost)
covs = np.identity(numCovarsCost)
Z = np.random.multivariate_normal(means, covs, numAgents)
# Construct level indicators for outcomes and choices.
Y1_level = np.dot(Y1_beta, X.T)
Y0_level = np.dot(Y0_beta, X.T)
D_level = np.dot(D_gamma, Z.T)
# Simulate unobservables from the model.
means = np.tile(0.0, 3)
vars_ = [U1_var, U0_var, V_var]
covs = np.diag(vars_)
covs[0,2] = U1V_cov
covs[2,0] = covs[0,2]
covs[1,2] = U0V_cov
covs[2,1] = covs[1,2]
U = np.random.multivariate_normal(means, covs, numAgents)
# Simulate individual outcomes and choices.
Y1 = np.tile(np.nan, (numAgents))
Y0 = np.tile(np.nan, (numAgents))
Y = np.tile(np.nan, (numAgents))
D = np.tile(np.nan, (numAgents))
for i in range(numAgents):
# Distribute unobservables.
U1 = U[i,0]
U0 = U[i,1]
V = U[i,2]
# Decision Rule.
expectedBenefits = Y1_level[i] - Y0_level[i]
cost = D_level[i] + V
D[i] = np.float((expectedBenefits - cost > 0))
# Potential outcomes.
Y1[i] = Y1_level[i] + U1
Y0[i] = Y0_level[i] + U0
# Observed outcomes.
Y[i] = D[i]*Y1[i] + (1.0 - D[i])*Y0[i]
# Check quality of simulated sample.
assert (np.all(np.isfinite(Y1)))
assert (np.all(np.isfinite(Y0)))
assert (np.all(np.isfinite(Y)))
assert (np.all(np.isfinite(D)))
assert (Y1.shape == (numAgents, ))
assert (Y0.shape == (numAgents, ))
assert (Y.shape == (numAgents, ))
assert (D.shape == (numAgents, ))
assert (Y1.dtype == 'float')
assert (Y0.dtype == 'float')
assert (Y.dtype == 'float')
assert (D.dtype == 'float')
assert ((D.all() in [1.0, 0.0]))
# Export sample to *.txt file for further processing.
np.savetxt(fileName, np.column_stack((Y, D, X, Z)), fmt= '%8.3f')
def estimate(outputfile = False):
''' Public interface to request an estimation of the Generalized
Roy Model.
'''
print "Estimating."
# Checks.
assert (os.path.exists('grmInit.ini'))
# Process initialization file.
initDict = grmReader.read()
# Checks.
assert (os.path.exists(initDict['fileName']))
# Process initFile.
_initializeLogging()
#
# Distribute useful covariates.
#
numAgents = initDict['numAgents']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
maxiter = initDict['maxiter']
#
# Construct auxiliary objects.
#
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
#
# Read and check dataset, distribute entries.
#
data = np.genfromtxt(initDict['fileName'], dtype = 'float')
assert (_checkData(data, numAgents, numCovarsOut, numCovarsCost) == True)
Y = data[:,0]
D = data[:,1]
X = data[:,2:(numCovarsOut + 2)]
Z = data[:,-numCovarsCost:]
#
# Maximization Script.
#
# Construct starting values.
startVals = np.concatenate((Y1_beta, Y0_beta, D_gamma,
[U1_var], [U0_var], [U1V_rho], [U0V_rho]))
# Run maximization.
sys.stdout = open('grmLogging.txt', 'a')
rslt = fmin_bfgs(_maxAlgorithmInterface, startVals, \
args = (Y, D, X, Z), maxiter = maxiter, \
full_output = False)
sys.stdout = sys.__stdout__
# Output
rslt = _distributeEvaluationValues(rslt, numCovarsOut, True)
rslt['Y1_beta'] = rslt['Y1_beta'].tolist()
rslt['Y0_beta'] = rslt['Y0_beta'].tolist()
rslt['D_gamma'] = rslt['D_gamma'].tolist()
if outputfile is not False:
with open(outputfile, 'w') as file_:
json.dump(rslt, file_)
print "Estimates saved to \'{}\'.".format(outputfile)
return rslt
def estimate():
""" Public interface to request an estimation of the Generalized
Roy Model.
"""
# Checks.
assert os.path.exists("grmInit.ini")
# Process initialization file.
initDict = grmReader.read()
# Checks.
assert os.path.exists(initDict["fileName"])
# Process initFile.
_initializeLogging()
""" Distribute useful covariates.
"""
numAgents = initDict["numAgents"]
Y1_beta = np.array(initDict["Y1_beta"])
Y0_beta = np.array(initDict["Y0_beta"])
D_gamma = np.array(initDict["D_gamma"])
U1_var = initDict["U1_var"]
U0_var = initDict["U0_var"]
U1V_rho = initDict["U1V_rho"]
U0V_rho = initDict["U0V_rho"]
maxiter = initDict["maxiter"]
""" Construct auxiliary objects.
"""
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
""" Read and check dataset, distribute entries.
"""
data = np.genfromtxt(initDict["fileName"], dtype="float")
assert _checkData(data, numAgents, numCovarsOut, numCovarsCost) == True
Y = data[:, 0]
D = data[:, 1]
X = data[:, 2 : (numCovarsOut + 2)]
Z = data[:, -numCovarsCost:]
""" Maximization Script.
"""
# Construct starting values.
startVals = np.concatenate((Y1_beta, Y0_beta, D_gamma, [U1_var], [U0_var], [U1V_rho], [U0V_rho]))
# Run maximization.
sys.stdout = open("grmLogging.txt", "a")
rslt = fmin_bfgs(_maxAlgorihtmInterface, startVals, args=(Y, D, X, Z), maxiter=maxiter, full_output=True)
sys.stdout = sys.__stdout__
# Construct dictionary with results.
rslt = _distributeEvaluationValues(rslt, numCovarsOut, True)
# Write out the *.json file.
with open("grmRslt.json", "w") as file_:
json.dump(initDict, file_)
def simulate(numAgents):
''' Simulate data generation process of the Generalized Roy Model.
'''
# Process initFile.
initDict = grmReader.read()
''' Distribute parametrization and (limited) type conversions.
'''
#Turning off numAgents and fileName to let the ParallelRun do the job
#numAgents = initDict['numAgents']
#fileName = initDict['fileName']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
V_var = initDict['V_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
randomSeed = initDict['randomSeed']
''' Set random seed
'''
np.random.seed(randomSeed)
''' Construct auxiliary objects.
'''
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
U1V_cov = U1V_rho*np.sqrt(U1_var)*np.sqrt(V_var)
U0V_cov = U0V_rho*np.sqrt(U0_var)*np.sqrt(V_var)
''' Simulate observable agent characteristics.
'''
means = np.tile(0.0, numCovarsOut)
covs = np.identity(numCovarsOut)
X = np.random.multivariate_normal(means, covs, numAgents)
X[:,0] = 1.0
means = np.tile(0.0, numCovarsCost)
covs = np.identity(numCovarsCost)
Z = np.random.multivariate_normal(means, covs, numAgents)
''' Construct level indicators for outcomes and choices.
'''
Y1_level = np.dot(Y1_beta, X.T)
Y0_level = np.dot(Y0_beta, X.T)
D_level = np.dot(D_gamma, Z.T)
''' Simulate unobservables from the model.
'''
means = np.tile(0.0, 3)
vars_ = [U1_var, U0_var, V_var]
covs = np.diag(vars_)
covs[0,2] = U1V_cov
covs[2,0] = covs[0,2]
covs[1,2] = U0V_cov
covs[2,1] = covs[1,2]
U = np.random.multivariate_normal(means, covs, numAgents)
''' Simulate individual outcomes and choices.
'''
Y1 = np.tile(np.nan, (numAgents))
Y0 = np.tile(np.nan, (numAgents))
Y = np.tile(np.nan, (numAgents))
D = np.tile(np.nan, (numAgents))
for i in range(numAgents):
# Distribute unobservables.
U1 = U[i,0]
U0 = U[i,1]
V = U[i,2]
# Decision Rule.
expectedBenefits = Y1_level[i] - Y0_level[i]
cost = D_level[i] + V
D[i] = np.float((expectedBenefits - cost > 0))
# Potential outcomes.
Y1[i] = Y1_level[i] + U1
Y0[i] = Y0_level[i] + U0
# Observed outcomes.
Y[i] = D[i]*Y1[i] + (1.0 - D[i])*Y0[i]
''' Calculating the treatment effects and exporting
'''
ATE = np.sum(Y1-Y0)/np.size(Y1-Y0)
ATT = np.sum((Y1-Y0)[D==1])/np.size((Y1-Y0)[D==1])
ATUT = np.sum((Y1-Y0)[D==0])/np.size((Y1-Y0)[D==0])
#Deactivating the output below, since we'll do it in ParallelRun
#np.savetxt('Treatments.txt', np.column_stack((ATE,ATT,ATUT)), fmt= '%8.3f')
''' Check quality of simulated sample.
'''
assert (np.all(np.isfinite(Y1)))
assert (np.all(np.isfinite(Y0)))
assert (np.all(np.isfinite(Y)))
assert (np.all(np.isfinite(D)))
assert (Y1.shape == (numAgents, ))
assert (Y0.shape == (numAgents, ))
assert (Y.shape == (numAgents, ))
assert (D.shape == (numAgents, ))
assert (Y1.dtype == 'float')
assert (Y0.dtype == 'float')
assert (Y.dtype == 'float')
assert (D.dtype == 'float')
assert ((D.all() in [1.0, 0.0]))
assert (ATE.dtype == 'float')
assert (ATT.dtype == 'float')
assert (ATUT.dtype == 'float')
''' Non longer export sample to *.txt file for further processing.
Instead, output to a variable to be gathered by ParallelRun
'''
variablesOutput = np.column_stack( (Y, D, X, Z) )
treatmentsOutput = np.column_stack( (ATE, ATT, ATUT) )
return {'variables':variablesOutput, 'treatments': treatmentsOutput}
def simulate():
''' Simulate data generation process of the Generalized Roy Model.
'''
# Process initFile.
initDict = grmReader.read()
''' Distribute parametrization and (limited) type conversions.
'''
numAgents = initDict['numAgents']
fileName = initDict['fileName']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
V_var = initDict['V_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
randomSeed = initDict['randomSeed']
''' Set random seed
'''
np.random.seed(randomSeed)
''' Construct auxiliary objects.
'''
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
U1V_cov = U1V_rho*np.sqrt(U1_var)*np.sqrt(V_var)
U0V_cov = U0V_rho*np.sqrt(U0_var)*np.sqrt(V_var)
''' Simulate observable agent characteristics.
'''
means = np.tile(0.0, numCovarsOut)
covs = np.identity(numCovarsOut)
X = np.random.multivariate_normal(means, covs, numAgents)
X[:,0] = 1.0
means = np.tile(0.0, numCovarsCost)
covs = np.identity(numCovarsCost)
Z = np.random.multivariate_normal(means, covs, numAgents)
''' Construct level indicators for outcomes and choices.
'''
Y1_level = np.dot(Y1_beta, X.T)
Y0_level = np.dot(Y0_beta, X.T)
D_level = np.dot(D_gamma, Z.T)
''' Simulate unobservables from the model.
'''
means = np.tile(0.0, 3)
vars_ = [U1_var, U0_var, V_var]
covs = np.diag(vars_)
covs[0,2] = U1V_cov
covs[2,0] = covs[0,2]
covs[1,2] = U0V_cov
covs[2,1] = covs[1,2]
U = np.random.multivariate_normal(means, covs, numAgents)
''' Simulate individual outcomes and choices.
'''
Y1 = np.tile(np.nan, (numAgents))
Y0 = np.tile(np.nan, (numAgents))
Y = np.tile(np.nan, (numAgents))
D = np.tile(np.nan, (numAgents))
for i in range(numAgents):
# Distribute unobservables.
U1 = U[i,0]
U0 = U[i,1]
V = U[i,2]
# Decision Rule.
expectedBenefits = Y1_level[i] - Y0_level[i]
cost = D_level[i] + V
D[i] = np.float((expectedBenefits - cost > 0))
# Potential outcomes.
Y1[i] = Y1_level[i] + U1
Y0[i] = Y0_level[i] + U0
# Observed outcomes.
Y[i] = D[i]*Y1[i] + (1.0 - D[i])*Y0[i]
''' Check quality of simulated sample.
'''
assert (np.all(np.isfinite(Y1)))
assert (np.all(np.isfinite(Y0)))
assert (np.all(np.isfinite(Y)))
assert (np.all(np.isfinite(D)))
assert (Y1.shape == (numAgents, ))
assert (Y0.shape == (numAgents, ))
assert (Y.shape == (numAgents, ))
assert (D.shape == (numAgents, ))
assert (Y1.dtype == 'float')
assert (Y0.dtype == 'float')
assert (Y.dtype == 'float')
assert (D.dtype == 'float')
assert ((D.all() in [1.0, 0.0]))
''' Export sample to *.txt file for further processing.
'''
np.savetxt(fileName, np.column_stack((Y, D, X, Z)), fmt= '%8.3f')
'''
grmToolbox.simulate()
''' Estimation.
'''
rslt=grmToolbox.estimate()
Y1_beta = (rslt['Y1_beta'])
Y0_beta = (rslt['Y0_beta'])
U1_var = rslt['U1_var']
U0_var = rslt['U0_var']
# Get the X's from the data
# Checks.
assert (os.path.exists('grmInit.ini'))
# Process initialization file.
initDict = grmReader.read()
data = numpy.genfromtxt(initDict['fileName'], dtype = 'float')
trash = numpy.array(initDict['Y1_beta'])
numCovarsOut = trash.shape[0]
X = data[:,2:(numCovarsOut + 2)]
D = data[:,1]
N_type1=sum(D)
index_0=numpy.where(D==0)[0]
index_1=numpy.where(D==1)[0]
# Now Create Y1 and Y0 without the errors; we will generate the errors later
# using Parallel Computing
D1_wo_errors_Y1=numpy.dot(X[index_1,:],Y1_beta)
D1_wo_errors_Y1_avg=numpy.mean(D1_wo_errors_Y1)
D1_wo_errors_Y0=numpy.dot(X[index_1,:],Y0_beta)
def estimate():
''' Public interface to request an estimation of the Generalized
Roy Model.
'''
# Checks.
assert (os.path.exists('grmInit.ini'))
# Process initialization file.
initDict = grmReader.read()
# Checks.
assert (os.path.exists(initDict['fileName']))
# Process initFile.
_initializeLogging()
''' Distribute useful covariates.
'''
numAgents = initDict['numAgents']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
maxiter = initDict['maxiter']
''' Construct auxiliary objects.
'''
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
''' Read and check dataset, distribute entries.
'''
data = np.genfromtxt(initDict['fileName'], dtype = 'float')
assert (_checkData(data, numAgents, numCovarsOut, numCovarsCost) == True)
Y = data[:,0]
D = data[:,1]
X = data[:,2:(numCovarsOut + 2)]
Z = data[:,-numCovarsCost:]
''' Maximization Script.
'''
# Construct starting values.
startVals = np.concatenate((Y1_beta, Y0_beta, D_gamma,
[U1_var], [U0_var], [U1V_rho], [U0V_rho]))
# Run maximization.
sys.stdout = open('grmLogging.txt', 'a')
rslt = fmin_bfgs(_maxAlgorihtmInterface, startVals, \
args = (Y, D, X, Z), maxiter = maxiter, \
full_output = True)
sys.stdout = sys.__stdout__
# Construct dictionary with results.
rslt = _distributeEvaluationValues(rslt[0], numCovarsOut, True)
return rslt
def evaluate():
# read grmRslt.json and get the dictionary
para=_getpara()
# get the estimated parameters from the dictionary
Y1_beta=para['Y1_beta']
Y0_beta=para['Y0_beta']
D_gamma=para['D_gamma']
U1_var=para['U1_var']
U0_var=para['U0_var']
U1V_rho=para['U1V_rho']
U0V_rho=para['U0V_rho']
# normalization
V_var=1
# calculate the covariance between U1 and V
U1V_cov=U1V_rho*np.sqrt(U1_var)*np.sqrt(V_var)
# calculate covariance between U0 and V
U0V_cov=U0V_rho*np.sqrt(U0_var)*np.sqrt(V_var)
# get the data from the .dat file
data_=_getdata()
# get numbAgents, numCovarsOut, numCovarsCost and randomSeed from the ini.file.
initDict = grmReader.read()
numAgents = initDict['numAgents']
Y1_beta_=initDict['Y1_beta']
numCovarsOut = np.array(Y1_beta_).shape[0]
D_gamma_=initDict['D_gamma']
numCovarsCost = np.array(D_gamma_).shape[0]
randomSeed=initDict['randomSeed']
#set random seed
np.random.seed(randomSeed)
# get the simulated X-covariates
X=data_[:,2:(numCovarsOut + 2)]
# get the simulated Z-covariates
Z=data_[:,-numCovarsCost:]
# calculate the level of Y_1 by using the estimated Y1_beta and simulated X
Y1_level=np.dot(Y1_beta,X.T)
# calculate the level of Y_0 by using the estimated Y0_beta and simulated X
Y0_level=np.dot(Y0_beta,X.T)
# calculate the level of D by using the estimated D_gamma and simulated Z
D_level=np.dot(D_gamma,Z.T)
# simulate the unobservables based on the estimated distributions
var_=[U1_var, U0_var, V_var]
cov=np.diag(var_)
cov[0,2]=U1V_cov
cov[2,0]=cov[0,2]
cov[1,2]=U0V_cov
cov[2,1]=cov[1,2]
U = np.random.multivariate_normal(np.tile(0.0,3), cov, numAgents)
U1=U[:,0]
U0=U[:,1]
V=U[:,2]
# simulate people's decisions
D = np.array((Y1_level-Y0_level+U1-U0-D_level-V) > 0)
# get the number of people who are treated (D=1)
numTreated=sum(D)
# get the number of people who are untreated (D=0)
numUntreated=numAgents-numTreated
# checks
assert (_checkdata(X,Y1_level,Y0_level,numAgents)==True)
'''calculate ATE
'''
ATE = (sum(Y1_level)-sum(Y0_level)+sum(U1)-sum(U0))/numAgents
'''calculate ATT
'''
# create an index indicating people who are treated (D=1)
index_tr=np.where(D==1)[0]
# get the X-covariates for people who are treated (D=1)
X_tr=X[index_tr,:]
# get U1 for people who are treated (D=1)
U1_tr=U1[index_tr,:]
# get U0 for people who are treated (D=1)
U0_tr=U0[index_tr,:]
# calculate the level of Y_1 for the treated agents (D=1)
Y1_level_tr=np.dot(Y1_beta,X_tr.T)
# calculate the level of Y_0 for the treated agents (D=1)
Y0_level_tr=np.dot(Y0_beta,X_tr.T)
# checks
assert (_checkdata(X_tr,Y1_level_tr,Y0_level_tr,numTreated)==True)
# calculate ATT
ATT = (sum(Y1_level_tr)-sum(Y0_level_tr)+sum(U1_tr)-sum(U0_tr))/numTreated
''' calculate ATU
'''
# create an index indicating people who are untreated (D=0)
index_utr=np.where(D==0)[0]
# get the X-covariates for people who are untreated (D=0)
X_utr=X[index_utr,:]
# get U1 for people who are untreated (D=0)
U1_utr=U1[index_utr,:]
# get U0 for people who are untreated (D=0)
U0_utr=U0[index_utr,:]
# calculate the level of Y_1 for the untreated agents (D=0)
Y1_level_utr=np.dot(Y1_beta,X_utr.T)
# calculate the level of Y_1 for the untreated agents (D=0)
Y0_level_utr=np.dot(Y0_beta,X_utr.T)
# checks
assert (_checkdata(X_utr,Y1_level_utr,Y0_level_utr,numUntreated)==True)
# calculate ATU
ATU = (sum(Y1_level_utr)-sum(Y0_level_utr)+sum(U1_utr)-sum(U0_utr))/numUntreated
#MPI
TreatmentEffects = np.array(comm.gather([ATE,ATT,ATU],root=0))
if rank==0:
TreatmentEffects = np.mean(TreatmentEffects, axis=0)
Treatments={}
Treatments['ATE'] = TreatmentEffects[0]
Treatments['ATT'] = TreatmentEffects[1]
Treatments['ATU']= TreatmentEffects[2]
print "ATE = %s" % Treatments['ATE']
print "ATT = %s" % Treatments['ATT']
print "ATU = %s" % Treatments['ATU']
def estimate():
''' Public interface to request an estimation of the Generalized
Roy Model.
'''
# Checks.
assert (os.path.exists('grmInit.ini'))
# Process initialization file.
initDict = grmReader.read()
# Checks.
assert (os.path.exists(initDict['fileName']))
# Process initFile.
_initializeLogging()
''' Distribute useful covariates.
'''
numAgents = initDict['numAgents']
Y1_beta = np.array(initDict['Y1_beta'])
Y0_beta = np.array(initDict['Y0_beta'])
D_gamma = np.array(initDict['D_gamma'])
U1_var = initDict['U1_var']
U0_var = initDict['U0_var']
U1V_rho = initDict['U1V_rho']
U0V_rho = initDict['U0V_rho']
maxiter = initDict['maxiter']
''' Construct auxiliary objects.
'''
numCovarsOut = Y1_beta.shape[0]
numCovarsCost = D_gamma.shape[0]
''' Read and check dataset, distribute entries.
'''
data = np.genfromtxt(initDict['fileName'], dtype='float')
assert (_checkData(data, numAgents, numCovarsOut, numCovarsCost) == True)
Y = data[:, 0]
D = data[:, 1]
X = data[:, 2:(numCovarsOut + 2)]
Z = data[:, -numCovarsCost:]
''' Maximization Script.
'''
# Construct starting values.
startVals = np.concatenate(
(Y1_beta, Y0_beta, D_gamma, [U1_var], [U0_var], [U1V_rho], [U0V_rho]))
# Run maximization.
sys.stdout = open('grmLogging.txt', 'a')
rslt = fmin_bfgs(_maxAlgorihtmInterface, startVals, \
args = (Y, D, X, Z), maxiter = maxiter, \
full_output = True)
sys.stdout = sys.__stdout__
# Construct dictionary with results.
rslt = _distributeEvaluationValues(rslt, numCovarsOut, True)
# Write out the *.json file.
with open('grmRslt.json', 'w') as file_:
json.dump(initDict, file_)
