def main():
# Logger
path = os.path.dirname(os.path.realpath(__file__))
file = os.path.join(path, "MMA_FUNCTION.log")
logger = setup_logger(file)
logger.info("Started\n")
# Set numpy print options
np.set_printoptions(precision=4, formatter={'float': '{: 0.4f}'.format})
# Initial settings
m = 1
n = 1
epsimin = 0.0000001
eeen = np.ones((n, 1))
eeem = np.ones((m, 1))
zeron = np.zeros((n, 1))
zerom = np.zeros((m, 1))
xval = 1 * eeen
xold1 = xval.copy()
xold2 = xval.copy()
xmin = eeen.copy()
xmax = 100 * eeen
low = xmin.copy()
upp = xmax.copy()
move = 1.0
c = 1000 * eeem
d = eeem.copy()
a0 = 1
a = zerom.copy()
outeriter = 0
maxoutit = 20
kkttol = 0
# Calculate function values and gradients of the objective and constraints functions
if outeriter == 0:
f0val, df0dx, fval, dfdx = funct(xval, n, eeen, zeron)
innerit = 0
outvector1 = np.array([outeriter, innerit, f0val, fval])
outvector2 = xval.flatten()
# Log
logger.info("outvector1 = {}".format(outvector1))
logger.info("outvector2 = {}\n".format(outvector2))
# The iterations starts
kktnorm = kkttol + 10
outit = 0
while (kktnorm > kkttol) and (outit < maxoutit):
outit += 1
outeriter += 1
# The MMA subproblem is solved at the point xval:
xmma,ymma,zmma,lam,xsi,eta,mu,zet,s,low,upp = \
mmasub(m,n,outeriter,xval,xmin,xmax,xold1,xold2,f0val,df0dx,fval,dfdx,low,upp,a0,a,c,d,move)
# Some vectors are updated:
xold2 = xold1.copy()
xold1 = xval.copy()
xval = xmma.copy()
# Re-calculate function values and gradients of the objective and constraints functions
f0val, df0dx, fval, dfdx = funct(xval, n, eeen, zeron)
# The residual vector of the KKT conditions is calculated
residu,kktnorm,residumax = \
kktcheck(m,n,xmma,ymma,zmma,lam,xsi,eta,mu,zet,s,xmin,xmax,df0dx,fval,dfdx,a0,a,c,d)
outvector1 = np.array([outeriter, innerit, f0val, fval])
outvector2 = xval.flatten()
# Log
logger.info("outvector1 = {}".format(outvector1))
logger.info("outvector2 = {}".format(outvector2))
logger.info("kktnorm = {}\n".format(kktnorm))
# Final log
logger.info(" Finished")
def main():
# Logger
path = os.path.dirname(os.path.realpath(__file__))
file = os.path.join(path, "GCMMA_BEAM.log")
logger = setup_logger(file)
logger.info("Started\n")
# Set numpy print options
np.set_printoptions(precision=4, formatter={'float': '{: 0.4f}'.format})
# Beam initial settings
m = 1
n = 5
epsimin = 0.0000001
eeen = np.ones((n, 1))
eeem = np.ones((m, 1))
zeron = np.zeros((n, 1))
zerom = np.zeros((m, 1))
xval = 5 * eeen
xold1 = xval.copy()
xold2 = xval.copy()
xmin = eeen.copy()
xmax = 10 * eeen
low = xmin.copy()
upp = xmax.copy()
c = 1000 * eeem
d = eeem.copy()
a0 = 1
a = zerom.copy()
raa0 = 0.01
raa = 0.01 * eeem
raa0eps = 0.000001
raaeps = 0.000001 * eeem
outeriter = 0
maxoutit = 20
kkttol = 0
# Calculate function values and gradients of the objective and constraints functions
if outeriter == 0:
f0val, df0dx, fval, dfdx = beam2(xval)
innerit = 0
outvector1 = np.array([outeriter, innerit, f0val, fval])
outvector2 = xval.flatten()
# Log
logger.info("outvector1 = {}".format(outvector1))
logger.info("outvector2 = {}\n".format(outvector2))
# The iterations starts
kktnorm = kkttol + 10
outit = 0
while (kktnorm > kkttol) and (outit < maxoutit):
outit += 1
outeriter += 1
# The parameters low, upp, raa0 and raa are calculated:
low,upp,raa0,raa= \
asymp(outeriter,n,xval,xold1,xold2,xmin,xmax,low,upp,raa0,raa,raa0eps,raaeps,df0dx,dfdx)
# The MMA subproblem is solved at the point xval:
xmma,ymma,zmma,lam,xsi,eta,mu,zet,s,f0app,fapp= \
gcmmasub(m,n,iter,epsimin,xval,xmin,xmax,low,upp,raa0,raa,f0val,df0dx,fval,dfdx,a0,a,c,d)
# The user should now calculate function values (no gradients) of the objective- and constraint
# functions at the point xmma ( = the optimal solution of the subproblem).
f0valnew, fvalnew = beam1(xmma)
# It is checked if the approximations are conservative:
conserv = concheck(m, epsimin, f0app, f0valnew, fapp, fvalnew)
# While the approximations are non-conservative (conserv=0), repeated inner iterations are made:
innerit = 0
if conserv == 0:
while conserv == 0 and innerit <= 15:
innerit += 1
# New values on the parameters raa0 and raa are calculated:
raa0,raa = raaupdate(xmma,xval,xmin,xmax,low,upp,f0valnew,fvalnew,f0app,fapp,raa0, \
raa,raa0eps,raaeps,epsimin)
# The GCMMA subproblem is solved with these new raa0 and raa:
xmma,ymma,zmma,lam,xsi,eta,mu,zet,s,f0app,fapp = gcmmasub(m,n,iter,epsimin,xval,xmin, \
xmax,low,upp,raa0,raa,f0val,df0dx,fval,dfdx,a0,a,c,d)
# The user should now calculate function values (no gradients) of the objective- and
# constraint functions at the point xmma ( = the optimal solution of the subproblem).
f0valnew, fvalnew = beam1(xmma)
# It is checked if the approximations have become conservative:
conserv = concheck(m, epsimin, f0app, f0valnew, fapp, fvalnew)
# Some vectors are updated:
xold2 = xold1.copy()
xold1 = xval.copy()
xval = xmma.copy()
# Re-calculate function values and gradients of the objective and constraints functions
f0val, df0dx, fval, dfdx = beam2(xval)
# The residual vector of the KKT conditions is calculated
residu,kktnorm,residumax = \
kktcheck(m,n,xmma,ymma,zmma,lam,xsi,eta,mu,zet,s,xmin,xmax,df0dx,fval,dfdx,a0,a,c,d)
outvector1 = np.array([outeriter, innerit, f0val, fval])
outvector2 = xval.flatten()
# Log
logger.info("outvector1 = {}".format(outvector1))
logger.info("outvector2 = {}".format(outvector2))
logger.info("kktnorm = {}\n".format(kktnorm))
# Final log
logger.info("Finished")