def execute(self):
""" Run the model by invoking run() on the workflow. """
self.iter_count = 0
# get initial values of design vars
dvinit = [eval(translate_expr(a, self)) for a in self.design_vars]
self.xmin = fmin_cobyla(self._driven_func, dvinit, self._consfuncts,
consargs=(),
rhobeg=self.rhobeg, rhoend=self.rhoend,
iprint=self.iprint, maxfun=self.maxiters)
def test_simple(self):
function = lambda x: x[0]**2 + abs(x[1])**3
con1 = lambda x: x[0]**2 + x[1]**2 - 25
con2 = lambda x: -con1(x)
x = co.fmin_cobyla(function, [4.95,0.66], [con1, con2], rhobeg=1,
rhoend=1e-5, iprint=0, maxfun=100)
x1 = 2.0/3
x0 = math.sqrt(25-x1*x1)
assert_almost_equal(x, [x0, x1], decimal=5)
def execute(self):
""" Run the model by invoking run() on the workflow. """
self.iter_count = 0
# get initial values of design vars
dvinit = [eval(translate_expr(a, self)) for a in self.design_vars]
self.xmin = fmin_cobyla(self._driven_func,
dvinit,
self._consfuncts,
consargs=(),
rhobeg=self.rhobeg,
rhoend=self.rhoend,
iprint=self.iprint,
maxfun=self.maxiters)
def test_simple(self):
function = lambda x: x[0]**2 + abs(x[1])**3
con1 = lambda x: x[0]**2 + x[1]**2 - 25
con2 = lambda x: -con1(x)
x = co.fmin_cobyla(function, [4.95, 0.66], [con1, con2],
rhobeg=1,
rhoend=1e-5,
iprint=0,
maxfun=100)
x1 = 2.0 / 3
x0 = math.sqrt(25 - x1 * x1)
assert_almost_equal(x, [x0, x1], decimal=5)
f = open('run_history.txt','a')
f.write('cw = {0}, cw length = {1}, cw beam length = {2}, sling length = {3}, range = {4}\n'.format(cw_dict['mass'],l2,l1,l4,r))
f.close()
'''
Since scipy optimizers can only minimize, multiply by -1
'''
return -1.0*r
def positive_cw_mass(x):
return x[cw_mass_index]
def positive_cw_beam_length(x):
return x[cw_beam_length_index]
def positive_cw_length(x):
return x[cw_length_index]
def positive_sling_length(x):
return x[sling_length_index]
if __name__ == '__main__':
if os.path.exists('./run_history.txt'):
os.remove('run_history.txt')
print co.fmin_cobyla(projectile_range,
x_init,
[positive_cw_mass,
positive_cw_beam_length,
positive_cw_length,
positive_sling_length],
rhobeg=0.1,
rhoend=0.001,
iprint=3,
maxfun=40)
def optimize(PS, figureOfMerit="mass_lbm", desVars=None,
findmin=1, MaxLoop=500, printLevel=1):
'''to find max, set findmin = 0'''
PS.firstEvaluateIfReqd()
global optFSys, _optLoopCount, _findmin, _constraintLoopCount
optFSys = PS
_optLoopCount = 0
_constraintLoopCount = 0
_findmin = findmin # use global flag in objective function
PS.figureOfMerit = figureOfMerit
PS.optDesVars = []
for dvStr in desVars:
dv = PS.desVarDict[dvStr]
if PS.hasFeasibleControlVar( dv.name ):
print 'WARNING... %s is a feasible design variable\n It can NOT be used as an optimization variable.'%dvStr
pass
else:
PS.optDesVars.append( dvStr )
#PS.summFile.write('\nPRIOR TO OPTIMIZATION\n')
if _findmin:
mLabel = 'MINIMIZE'
else:
mLabel = 'MAXIMIZE'
if printLevel:
saveOptVarSummary(PS, PS.optDesVars,'\nPRIOR TO %s OPTIMIZATION\n'%mLabel)
ndv = len( PS.optDesVars )
ncon = 0
if ndv <= 0:
print "ERROR... there are no legal design variables for optimization"
return
xlow=[]
xhigh=[]
xinit = []
# get limits of design variables
i = 0
for dvStr in PS.optDesVars:
dv = PS.desVarDict[dvStr]
xlow.append( dv.minVal / dv.scaleFactor )
xhigh.append( dv.maxVal / dv.scaleFactor )
xinit.append( PS.getDesignVar( dvStr) / dv.scaleFactor )
i += 1
# count any result variables that might be constraints
PS.constraintList[:] = []
for key,rv in PS.resultVarDict.items():
# only handle inequality constraints, equality constraints handled by feasibility variables
if rv.hasLowConstraint() :
PS.constraintList.append( ['>',key] )
ncon += 1
if rv.hasHighConstraint():
PS.constraintList.append( ['<',key] )
ncon += 1
#xOut = mdot.findminormax(findmin,ndv,ncon,MaxLoop, xlow,xhigh,xinit,objAndConstraints)
# set up constraint calling functions
cons = []
# first do bounds on x array
def makeLo(n):
return lambda x: x[n] - xlow[n]
def makeHi(n):
return lambda x: xhigh[n] - x[n]
for n in range( len(xlow) ):
conLo = makeLo(n)
conHi = makeHi(n)
cons.append( conLo )
cons.append( conHi )
# then do any constraints on result variables
def makeCon(n):
return lambda x: constraintFunc(x, n)
for n in range( len(PS.constraintList) ):
con = makeCon(n)
cons.append( con )
# rhobeg can be 1.0 because x values are scaled.
xOut = fmin_cobyla(objFunction, xinit, cons, rhobeg=1.0, rhoend=1e-6,
iprint=cast.intDammit(printLevel), maxfun=MaxLoop)
if printLevel:
print
print " ==> OPTIMIZATION (Loops = %i)"%_optLoopCount, "(Max = %g)"%MaxLoop,
print " (%i constraint loops)\n"%_constraintLoopCount
for i in range( len(PS.optDesVars) ):
dvStr = PS.optDesVars[i]
dv = PS.desVarDict[ dvStr ]
dvVal = xOut[i] * dv.scaleFactor
if printLevel:
print 'optimum',dvStr,'=',dvVal
#PS.summFile.write( '\noptimum %10s = %g'%(dvStr, dvVal) )
if printLevel:
print '\nscaled desVars',PS.optDesVars,'=',xOut[:ndv]
print 'gives',figureOfMerit,'=',PS.getResultVar(figureOfMerit),'\n'
#PS.summFile.write('\n\ngives '+ str(figureOfMerit)+\
# ' = '+ str(PS.getResultVar(figureOfMerit))+'\n')
for i in range(ndv):
dvStr = PS.optDesVars[i]
dv = PS.desVarDict[ dvStr ]
dvVal = xOut[i] * dv.scaleFactor
PS.setDesignVar( dvStr, dvVal)
PS.evaluate()
#PS.summFile.write('\nAFTER OPTIMIZATION\n')
if printLevel:
saveOptVarSummary(PS, PS.optDesVars,'\nAFTER %s OPTIMIZATION\n'%mLabel)
def optimize(PS,
figureOfMerit="mass_lbm",
desVars=None,
findmin=1,
MaxLoop=500,
printLevel=1):
'''to find max, set findmin = 0'''
PS.firstEvaluateIfReqd()
global optFSys, _optLoopCount, _findmin, _constraintLoopCount
optFSys = PS
_optLoopCount = 0
_constraintLoopCount = 0
_findmin = findmin # use global flag in objective function
PS.figureOfMerit = figureOfMerit
PS.optDesVars = []
for dvStr in desVars:
dv = PS.desVarDict[dvStr]
if PS.hasFeasibleControlVar(dv.name):
print 'WARNING... %s is a feasible design variable\n It can NOT be used as an optimization variable.' % dvStr
pass
else:
PS.optDesVars.append(dvStr)
#PS.summFile.write('\nPRIOR TO OPTIMIZATION\n')
if _findmin:
mLabel = 'MINIMIZE'
else:
mLabel = 'MAXIMIZE'
if printLevel:
saveOptVarSummary(PS, PS.optDesVars,
'\nPRIOR TO %s OPTIMIZATION\n' % mLabel)
ndv = len(PS.optDesVars)
ncon = 0
if ndv <= 0:
print "ERROR... there are no legal design variables for optimization"
return
xlow = []
xhigh = []
xinit = []
# get limits of design variables
i = 0
for dvStr in PS.optDesVars:
dv = PS.desVarDict[dvStr]
xlow.append(dv.minVal / dv.scaleFactor)
xhigh.append(dv.maxVal / dv.scaleFactor)
xinit.append(PS.getDesignVar(dvStr) / dv.scaleFactor)
i += 1
# count any result variables that might be constraints
PS.constraintList[:] = []
for key, rv in PS.resultVarDict.items():
# only handle inequality constraints, equality constraints handled by feasibility variables
if rv.hasLowConstraint():
PS.constraintList.append(['>', key])
ncon += 1
if rv.hasHighConstraint():
PS.constraintList.append(['<', key])
ncon += 1
#xOut = mdot.findminormax(findmin,ndv,ncon,MaxLoop, xlow,xhigh,xinit,objAndConstraints)
# set up constraint calling functions
cons = []
# first do bounds on x array
def makeLo(n):
return lambda x: x[n] - xlow[n]
def makeHi(n):
return lambda x: xhigh[n] - x[n]
for n in range(len(xlow)):
conLo = makeLo(n)
conHi = makeHi(n)
cons.append(conLo)
cons.append(conHi)
# then do any constraints on result variables
def makeCon(n):
return lambda x: constraintFunc(x, n)
for n in range(len(PS.constraintList)):
con = makeCon(n)
cons.append(con)
# rhobeg can be 1.0 because x values are scaled.
xOut = fmin_cobyla(objFunction,
xinit,
cons,
rhobeg=1.0,
rhoend=1e-6,
iprint=cast.intDammit(printLevel),
maxfun=MaxLoop)
if printLevel:
print
print " ==> OPTIMIZATION (Loops = %i)" % _optLoopCount, "(Max = %g)" % MaxLoop,
print " (%i constraint loops)\n" % _constraintLoopCount
for i in range(len(PS.optDesVars)):
dvStr = PS.optDesVars[i]
dv = PS.desVarDict[dvStr]
dvVal = xOut[i] * dv.scaleFactor
if printLevel:
print 'optimum', dvStr, '=', dvVal
#PS.summFile.write( '\noptimum %10s = %g'%(dvStr, dvVal) )
if printLevel:
print '\nscaled desVars', PS.optDesVars, '=', xOut[:ndv]
print 'gives', figureOfMerit, '=', PS.getResultVar(figureOfMerit), '\n'
#PS.summFile.write('\n\ngives '+ str(figureOfMerit)+\
# ' = '+ str(PS.getResultVar(figureOfMerit))+'\n')
for i in range(ndv):
dvStr = PS.optDesVars[i]
dv = PS.desVarDict[dvStr]
dvVal = xOut[i] * dv.scaleFactor
PS.setDesignVar(dvStr, dvVal)
PS.evaluate()
#PS.summFile.write('\nAFTER OPTIMIZATION\n')
if printLevel:
saveOptVarSummary(PS, PS.optDesVars,
'\nAFTER %s OPTIMIZATION\n' % mLabel)
def optimize_sizes_cobyla(size_bins):
likelihood_func = likelihood(size_bins)
initial_guess = array(initial_x(size_bins))
constraints = get_constraints(size_bins)
return fmin_cobyla(likelihood_func, initial_guess, constraints, rhobeg=0.1, rhoend=0.01)
def positive_cw_mass(x):
return x[cw_mass_index]
def positive_cw_beam_length(x):
return x[cw_beam_length_index]
def positive_cw_length(x):
return x[cw_length_index]
def positive_sling_length(x):
return x[sling_length_index]
if __name__ == '__main__':
if os.path.exists('./run_history.txt'):
os.remove('run_history.txt')
print co.fmin_cobyla(projectile_range,
x_init, [
positive_cw_mass, positive_cw_beam_length,
positive_cw_length, positive_sling_length
],
rhobeg=0.1,
rhoend=0.001,
iprint=3,
maxfun=40)