# Define a parameter optimization problem in relation to the
# FD algorithm.
from wp_declaration import FD
from opal import ModelStructure, ModelData, Model
from opal.Solvers import NOMAD
# Return the error measure.
def get_error(parameters, measures):
return sum(measures["ERROR"])
# Define parameter optimization problem.
data = ModelData(FD)
struct = ModelStructure(objective=get_error) # Unconstrained
model = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
NOMAD.set_parameter(name='DISPLAY_STATS',
value='%3dBBE %7.1eSOL %8.3eOBJ %5.2fTIME')
NOMAD.solve(blackbox=model)
def test_trunk_optimize():
from trunk_optimize import prob
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
NOMAD.solve(blackbox=prob)
numberOfList = 16
listLengthStep = 25
for listType in listTypes:
for k in range(numberOfList):
n = listLengthStep * (k + 1)
probName = str(listType) + '-' + str(n) + '-50'
problems.append(TestProblem(name=probName))
#SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=coopsort, problems=problems)
struct = ModelStructure(
objective=MeasureFunction(sum_time),
constraints=[], # Unconstrained
neighborhood=get_neighbors)
prob = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
if __name__ == '__main__':
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=100)
NOMAD.set_parameter(name='DISPLAY_DEGREE', value=4)
NOMAD.set_parameter(name='EXTENDED_POLL_TRIGGER', value='r0.1')
NOMAD.set_parameter(name='MAX_MESH_INDEX', value='2')
NOMAD.set_parameter(name='MODEL_SEARCH_OPTIMISTIC', value='no')
NOMAD.solve(blackbox=prob)
def test_trunk_optimize():
from trunk_optimize import prob
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
NOMAD.solve(blackbox=prob)
'TRIDIA',
'WOODS'
]
]
#SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=trunk,
problems=problems,
parameters=params,
measures=trunk.measures)
struct = ModelStructure(objective=MeasureFunction(sum_heval, additivity=1),
constraints=[
(None, MeasureFunction(get_error, addivity=1), 0)
]) # One constraint get_error(p) <= 0
prob = Model(modelData=data,
modelStructure=struct,
platform='SMP',
synchoronized=False,
interruptible=True)
# Solve parameter optimization problem.
if __name__ == '__main__':
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
NOMAD.solve(model=prob)
from opal.TestProblemCollections import CUTEr
from dfo_declaration import DFO
from opal.Solvers import NOMAD
def avg_time(parameters, measures):
n = len(measures['CPU'])
if n == 0:
return 0.0
return (sum(measures["CPU"]) + 0.0) / (n + 0.0)
# Select real parameters from DFO.
params = [par for par in DFO.parameters if par.is_real]
# Select tiny unconstrained HS problems.
problems = [prb for prb in CUTEr.HS if prb.nvar <= 5 and prb.ncon == 0]
print 'Working with parameters ', [par.name for par in params]
print 'Testing on problems ', [prb.name for prb in problems]
data = ModelData(DFO, problems, params)
structure = ModelStructure(objective=avg_time, constraints=[]) # Unconstrained
# Instantiate black-box solver.
model = Model(modelData=data, modelStructure=structure)
# Solve parameter optimization problem.
NOMAD.solve(blackbox=model)
constraints=[(None, sum_unsolvability, 0)])
# Instantiate black-box solver.
#from opal.Platforms import LSF
#LSF.set_parameter({'MAX_TASK':10})
from opal.Platforms import SunGrid
SunGrid.set_config('-q', 'all.q')
SunGrid.set_parameter({'MAX_TASK': 10})
model = Model(modelData=data, modelStructure=structure, platform=SunGrid)
if len(sys.argv) > 2: # The initial point is provided by external file
f = open(sys.argv[2])
paramValues = eval(f.read())
f.close()
model.initial_points = [] # Reset the initial point set
for tag in paramValues:
model.add_initial_point(paramValues[tag])
#print model.get_initial_points()
# Solve parameter optimization problem.
from opal.Solvers import NOMAD
#NOMAD.set_parameter(name='MAX_BB_EVAL', value=100)
#NOMAD.set_parameter(name='SCALING', value='scaling.txt')
NOMAD.set_parameter(name='INITIAL_MESH_SIZE', value='(0.05 5 5 1 4 0.05)')
#NOMAD.set_parameter(name='MIN_MESH_SIZE', value=1.0e-6)
NOMAD.set_parameter(name='MAX_MESH_INDEX', value=6)
NOMAD.set_parameter(name='DISPLAY_DEGREE', value=4)
NOMAD.solve(blackbox=model)
# Define parameter optimization problem.
data = ModelData(algorithm=coopsort, problems=problems)
surrogate_data = ModelData(algorithm=coopsort, problems=surrogate_problems)
struct = ModelStructure(
objective=MeasureFunction(sum_time),
constraints=[], # Unconstrained
neighborhood=get_neighbors)
prob = Model(modelData=data, modelStructure=struct)
surrogate_prob = Model(modelData=surrogate_data, modelStructure=struct)
# Solve parameter optimization problem.
if __name__ == '__main__':
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=100)
NOMAD.set_parameter(name='DISPLAY_DEGREE', value=1)
NOMAD.set_parameter(
name="DISPLAY_STATS",
value="%3dEVAL %3dBBE %4dSGTE [ %15lSOL, ] %8.2eBBO %4.2fTIME")
NOMAD.set_parameter(
name="STATS_FILE",
value=
"iterations.txt %3dEVAL %3dBBE %4dSGTE [ %15lSOL, ] %8.2eBBO %4.2fTIME"
)
#NOMAD.set_parameter(name='EXTENDED_POLL_TRIGGER', value='r0.1')
NOMAD.set_parameter(name='MAX_MESH_INDEX', value='2')
NOMAD.solve(blackbox=prob, surrogate=surrogate_prob)
tmpStr = encodedNumber[0:i] + s_2 + encodedNumber[i+3:]
neighbor = list(parameters)
neighbor[0] = int(tmpStr)
neighbors.append(neighbor)
return neighbors
problems = [TestProblem(name=str(N)) for N in [100, 1000, 10000]]
#SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=coopsort,
problems=problems)
struct = ModelStructure(objective=MeasureFunction(sum_time),
constraints=[], # Unconstrained
neighborhood=get_neighbors)
prob = Model(modelData=data,
modelStructure=struct
)
# Solve parameter optimization problem.
if __name__ == '__main__':
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
#print get_neighbors([401])
NOMAD.solve(blackbox=prob)
"NCB20B",
"NONDQUAR",
"POWER",
"SENSORS",
"SINQUAD",
"TESTQUAD",
"TRIDIA",
"WOODS",
]
]
# SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=trunk, problems=problems, parameters=params, measures=trunk.measures)
struct = ModelStructure(
objective=MeasureFunction(sum_heval, additivity=1), constraints=[(None, MeasureFunction(get_error, addivity=1), 0)]
) # One constraint get_error(p) <= 0
prob = Model(modelData=data, modelStructure=struct, platform="SMP", synchoronized=False, interruptible=True)
# Solve parameter optimization problem.
if __name__ == "__main__":
from opal.Solvers import NOMAD
NOMAD.set_parameter(name="MAX_BB_EVAL", value=10)
NOMAD.solve(model=prob)
val = sum(measures["FEVAL"])
return val
# Parameters being tuned and problem list.
par_names = ['eta1', 'eta2', 'gamma1', 'gamma2', 'gamma3']
params = [param for param in trunk.parameters if param.name in par_names]
problems = [
problem for problem in CUTEr if problem.name in [
'BDQRTIC', 'BROYDN7D', 'BRYBND', 'CURLY10', 'CURLY20', 'CURLY30',
'CRAGGLVY', 'DIXON3DQ', 'EIGENALS', 'FMINSRF2', 'FMINSURF', 'GENROSE',
'HIELOW', 'MANCINO', 'NCB20', 'NCB20B', 'NONDQUAR', 'NONDQUAR',
'POWER', 'SENSORS', 'SINQUAD', 'TESTQUAD', 'TRIDIA', 'WOODS'
]
]
# Define parameter optimization problem.
LSF.set_config(parameterName="-m", parameterValue='"lin01 lin02 lin03 lin04"')
LSF.set_config(parameterName="-q", parameterValue="fedora")
data = ModelData(algorithm=trunk,
problems=problems,
parameters=params,
platform=LSF)
struct = ModelStructure(objective=get_error, constraints=[]) # Unconstrained
blackbox = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
NOMAD.solve(model=blackbox)
# Instantiate black-box solver.
#from opal.Platforms import LSF
#LSF.set_parameter({'MAX_TASK':10})
from opal.Platforms import SunGrid
SunGrid.set_config('-q','all.q')
SunGrid.set_parameter({'MAX_TASK':10})
model = Model(modelData=data, modelStructure=structure, platform=SunGrid)
if len(sys.argv) > 2: # The initial point is provided by external file
f = open(sys.argv[2])
paramValues = eval(f.read())
f.close()
model.initial_points = [] # Reset the initial point set
for tag in paramValues:
model.add_initial_point(paramValues[tag])
#print model.get_initial_points()
# Solve parameter optimization problem.
from opal.Solvers import NOMAD
#NOMAD.set_parameter(name='MAX_BB_EVAL', value=100)
#NOMAD.set_parameter(name='SCALING', value='scaling.txt')
NOMAD.set_parameter(name='INITIAL_MESH_SIZE',
value='(0.05 5 5 1 4 0.05)')
#NOMAD.set_parameter(name='MIN_MESH_SIZE', value=1.0e-6)
NOMAD.set_parameter(name='MAX_MESH_INDEX', value=6)
NOMAD.set_parameter(name='DISPLAY_DEGREE', value=4)
NOMAD.solve(blackbox=model)
'EIGENALS',
'FMINSRF2',
'FMINSURF',
'GENROSE',
'HIELOW',
'MANCINO',
'NCB20',
'NCB20B',
'NONDQUAR',
'POWER',
'SENSORS',
'SINQUAD',
'TESTQUAD',
'TRIDIA',
'WOODS']]
# Define parameter optimization problem.
data = ModelData(algorithm=trunk,
problems=problems,
parameters=params,
platform=OPALMPI)
struct = ModelStructure(objective=get_error,
constraints=[]) # Unconstrained
model = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
NOMAD.set_parameter(name='MAX_BB_EVAL', value=500)
NOMAD.solve(blackbox=model)
from opal import ModelStructure, ModelData, Model
from opal.TestProblemCollections import CUTEr
from dfo_declaration import DFO
from opal.Solvers import NOMAD
def avg_time(parameters,measures):
n = len(measures['CPU'])
if n == 0:
return 0.0
return (sum(measures["CPU"]) + 0.0)/(n + 0.0)
# Select real parameters from DFO.
params = [par for par in DFO.parameters if par.is_real]
# Select tiny unconstrained HS problems.
problems = [prb for prb in CUTEr.HS if prb.nvar <= 5 and prb.ncon == 0]
print 'Working with parameters ', [par.name for par in params]
print 'Testing on problems ', [prb.name for prb in problems]
data = ModelData(DFO, problems, params)
structure = ModelStructure(objective=avg_time, constraints=[]) # Unconstrained
# Instantiate black-box solver.
model = Model(modelData=data, modelStructure=structure)
# Solve parameter optimization problem.
NOMAD.solve(blackbox=model)
"DIXON3DQ",
"EIGENALS",
"FMINSRF2",
"FMINSURF",
"GENROSE",
"HIELOW",
"MANCINO",
"NCB20",
"NCB20B",
"NONDQUAR",
"NONDQUAR",
"POWER",
"SENSORS",
"SINQUAD",
"TESTQUAD",
"TRIDIA",
"WOODS",
]
]
# Define parameter optimization problem.
LSF.set_config(parameterName="-m", parameterValue='"lin01 lin02 lin03 lin04"')
LSF.set_config(parameterName="-q", parameterValue="fedora")
data = ModelData(algorithm=trunk, problems=problems, parameters=params, platform=LSF)
struct = ModelStructure(objective=get_error, constraints=[]) # Unconstrained
blackbox = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
NOMAD.set_parameter(name="MAX_BB_EVAL", value=10)
NOMAD.solve(model=blackbox)
listTypes = [0, 5, 6, 7]
numberOfList = 16
listLengthStep = 25
for listType in listTypes:
for k in range(numberOfList):
n = listLengthStep * (k + 1)
probName = str(listType) + '-' + str(n) + '-50'
problems.append(TestProblem(name=probName))
#SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=coopsort, problems=problems)
struct = ModelStructure(objective=MeasureFunction(sum_time),
constraints=[], # Unconstrained
neighborhood=get_neighbors)
prob = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
if __name__ == '__main__':
from opal.Solvers import NOMAD
NOMAD.set_parameter(name='MAX_BB_EVAL', value=100)
NOMAD.set_parameter(name='DISPLAY_DEGREE', value=4)
NOMAD.set_parameter(name='EXTENDED_POLL_TRIGGER', value='r0.1')
NOMAD.set_parameter(name='MAX_MESH_INDEX', value='2')
NOMAD.set_parameter(name='MODEL_SEARCH_OPTIMISTIC', value='no')
NOMAD.solve(blackbox=prob)
import sys
# Define a parameter optimization problem in relation to the
# LAS algorithm.
from las_declaration import LAS
from opal import ModelStructure, ModelData, Model
from opal.Solvers import NOMAD
# Return the error measure.
def total_time(parameters, measures):
with open("measures", "a") as f:
f.write("parameters = %r, measures = %r\n" % (parameters, measures))
# Now with numerical integration, we want to minimize total time, not the time/rel
# return sum(map(float, measures["TIME"])) / sum(map(float, measures["RELATIONS"]))
return sum(map(float, measures["SIEVETIME"]))
# Define parameter optimization problem.
data = ModelData(LAS)
struct = ModelStructure(objective=total_time) # Unconstrained
model = Model(modelData=data, modelStructure=struct)
# Solve parameter optimization problem.
NOMAD.set_parameter(name='DISPLAY_STATS',
value='%3dBBE %7.1eSOL %8.3eOBJ %5.2fTIME %5.2SIEVETIME')
NOMAD.solve(blackbox=model)
surrogate_problems.append(TestProblem(name=probName))
# SMP.set_parameter(name='MAX_PROC', value=5);
# Define parameter optimization problem.
data = ModelData(algorithm=coopsort, problems=problems)
surrogate_data = ModelData(algorithm=coopsort, problems=surrogate_problems)
struct = ModelStructure(
objective=MeasureFunction(sum_time), constraints=[], neighborhood=get_neighbors # Unconstrained
)
prob = Model(modelData=data, modelStructure=struct)
surrogate_prob = Model(modelData=surrogate_data, modelStructure=struct)
# Solve parameter optimization problem.
if __name__ == "__main__":
from opal.Solvers import NOMAD
NOMAD.set_parameter(name="MAX_BB_EVAL", value=100)
NOMAD.set_parameter(name="DISPLAY_DEGREE", value=1)
NOMAD.set_parameter(name="DISPLAY_STATS", value="%3dEVAL %3dBBE %4dSGTE [ %15lSOL, ] %8.2eBBO %4.2fTIME")
NOMAD.set_parameter(
name="STATS_FILE", value="iterations.txt %3dEVAL %3dBBE %4dSGTE [ %15lSOL, ] %8.2eBBO %4.2fTIME"
)
# NOMAD.set_parameter(name='EXTENDED_POLL_TRIGGER', value='r0.1')
NOMAD.set_parameter(name="MAX_MESH_INDEX", value="2")
NOMAD.solve(blackbox=prob, surrogate=surrogate_prob)
'SENSORS',
'SINQUAD',
'TESTQUAD',
'TRIDIA',
'WOODS']]
# Define parameter optimization problem.
data = ModelData(algorithm=trunk,
problems=problems,
parameters=params)
struct = ModelStructure(objective=sum_heval,
constraints=[(None,get_error, 0)])
model = Model(modelData=data, modelStructure=struct)
# Define a surrogate
surr_data = ModelData(algorithm=trunk,
problems= [problem for problem in CUTEr \
if problem.name in ['BDQRTIC',
'BROYDN7D',
'BRYBND']],
parameters=params)
surr_struct = ModelStructure(objective=sum_heval,
constraints=[])
surr_model = Model(modelData=surr_data, modelStructure=surr_struct,
dataFile='surrogate.dat')
# Solve parameter optimization problem.
NOMAD.set_parameter(name='MAX_BB_EVAL', value=10)
NOMAD.solve(blackbox=model, surrogate=surr_model)
