def bayes_opt_test():
# print("hello World",flush=True)
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
print(ffield_.data[0][3][11])
print(ffield_.data[0][3][12])
print(ffield_.data[0][3][15])
print(ffield_.data[0][3][16])
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (0.5, 6.0), linkedParams=[[0, 4, 0]])
]
ffieldParams_.append(
ForceFieldParam((0, 3, 12), (25.0, 600.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 15), (1.0, 8.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 16), (1000.0, 8000.0), linkedParams=[[0, 4,
0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
lmpFile = LammpsFile("lammpsTest.lmp")
structToEnergyMap = {
"data.alpha_Si3N4_1.0": -303.46972,
"data.alpha_Si3N4_0.98": -301.88812,
"data.alpha_Si3N4_1.02": -301.83053,
"data.alpha_Si3N4_0.9": -264.85529,
"data.alpha_Si3N4_1.1": -274.56516
}
def ObjFunc(a, b, c, d):
paramList = [a, b, c, d]
return (objectiveFunction_test2(paramList, ffieldParams=ffieldParams_, ffield=ffield_, \
lmpFile=lmpFile, structToEnergyMap=structToEnergyMap)[0][0])
pbounds = {
'a': ffieldParams_[0].paramRange,
'b': ffieldParams_[1].paramRange,
'c': ffieldParams_[2].paramRange,
'd': ffieldParams_[3].paramRange
}
from bayes_opt import SequentialDomainReductionTransformer
boundsTransformer = SequentialDomainReductionTransformer()
optimizer = BayesianOptimization(f=ObjFunc, pbounds=pbounds)
optimizer.maximize(
init_points=1000,
n_iter=100,
)
print(optimizer.max)
print("or does it equal="+str(ObjFunc(optimizer.max['params']['a'],optimizer.max['params']['b'],\
optimizer.max['params']['c'],optimizer.max['params']['d'])))
print("at random point" + str(ObjFunc(2.057, 293.8, 4.088, 4512.0)))
def skoptTest():
# print("hello World",flush=True)
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
print(ffield_.data[0][3][11])
print(ffield_.data[0][3][12])
print(ffield_.data[0][3][15])
print(ffield_.data[0][3][16])
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (0.5, 6.0), linkedParams=[[0, 4, 0]])
]
ffieldParams_.append(
ForceFieldParam((0, 3, 12), (25.0, 600.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 15), (1.0, 8.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 16), (1000.0, 8000.0), linkedParams=[[0, 4,
0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
lmpFile = LammpsFile("lammpsTest.lmp")
structToEnergyMap = {
"data.alpha_Si3N4_1.0": -303.46972,
"data.alpha_Si3N4_0.98": -301.88812,
"data.alpha_Si3N4_1.02": -301.83053,
"data.alpha_Si3N4_0.9": -264.85529,
"data.alpha_Si3N4_1.1": -274.56516
}
def ObjFunc(paramList):
return -(objectiveFunction_test2(paramList, ffieldParams=ffieldParams_, ffield=ffield_, \
lmpFile=lmpFile, structToEnergyMap=structToEnergyMap)[0][0])
from skopt import gp_minimize
res = gp_minimize(
ObjFunc, # the function to minimize
paramRanges, # the bounds on each dimension of x
acq_func="EI", # the acquisition function
n_calls=100, # the number of evaluations of f
n_random_starts=100, # the number of random initialization points
#noise=0.1 ** 2, # the noise level (optional)
n_restarts_optimizer=9
) #number of restarts for the acquisition optimizer
#random_state=1234) # the random seed
print(res.x)
print("min=" + str(res.fun))
print("or does it equal=" + str(ObjFunc(res.x)))
def trainsetReadingTest():
# print("hello World",flush=True)
ffieldParams_ = []
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (1.5, 6.0), linkedParams=[[0, 4, 0]])
]
ffieldParams_.append(
ForceFieldParam((0, 3, 12), (25.0, 600.0), linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 3, 15), (1.0, 8.0), linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 3, 16), (1000.0, 8000.0), linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 7, 8), (-1.0, 1.0))) # , linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 7, 6), (1000.0, 1000000.0))) # , linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 7, 7), (1.0, 1000.0))) # , linkedParams=[[0, 4, 0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
originalValues = [
ffield_.data[0][param.loc[1]][param.loc[2]] for param in ffieldParams_
]
print(originalValues)
objList = readTrainsetFile("Si3N4_DATA/difftrainset.test",
singleMeanSquaredError, "Si3N4_DATA/")
def objWrapper(simpleParamList):
return -objectiveFunctionFromPartials(simpleParamList, ffieldParams_,
ffield_, objList)
normBOpt = normalizationWrapper(paramRanges,
objWrapper,
startingSamples=50)
normBOpt.learn(50)
bOpt = normBOpt.bOpt
print("best error prior to GD:")
print(max(normBOpt.bOpt.sampleResults.ravel()))
bestIndex = argmax(bOpt.sampleResults.ravel())
print("corresponding paramSet:")
print(
BoA.descaleSamples0to1(bOpt.sampleArray[bestIndex],
normBOpt.paramRanges))
x = bOpt.optimalParams()
x = BoA.descaleSamples0to1(x, normBOpt.paramRanges)
print("\n------------\n\nbest error after GD:")
print(objWrapper(x))
print("corresponding paramSet:")
print(x)
print("THE KNOWN BEST PARAMS:")
print(objWrapper(originalValues))
def quickTest():
# print("hello World",flush=True)
ffieldParams_ = []
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
print(ffield_.data[0][3][11])
print(ffield_.data[0][3][12])
print(ffield_.data[0][3][15])
print(ffield_.data[0][3][16])
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (0.5, 6.0), linkedParams=[[0, 4, 0]])
]
ffieldParams_.append(
ForceFieldParam((0, 3, 12), (25.0, 600.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 15), (1.0, 8.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 16), (1000.0, 8000.0), linkedParams=[[0, 4,
0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
lmpFile = LammpsFile("lammpsTest.lmp")
structToEnergyMap = {
"data.alpha_Si3N4_1.0": -303.46972,
"data.alpha_Si3N4_0.98": -301.88812,
"data.alpha_Si3N4_1.02": -301.83053,
"data.alpha_Si3N4_0.9": -264.85529,
"data.alpha_Si3N4_1.1": -274.56516
}
def ObjFunc(paramSetList):
return -objectiveFunction_test2(paramSetList, ffieldParams=ffieldParams_, ffield=ffield_, \
lmpFile=lmpFile, structToEnergyMap=structToEnergyMap)
x = [1.25744366, 35.77402639, 4.51140332, 1233.33779981]
import scipy.optimize
result = scipy.optimize.minimize(ObjFunc,
x,
method='L-BFGS-B',
bounds=paramRanges)
print(result.x)
print(result.fun)
def normalizationTest():
# print("hello World",flush=True)
ffieldParams_ = []
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (1.5, 6.0), linkedParams=[[0, 4, 0]])
]
ffieldParams_.append(
ForceFieldParam((0, 3, 12), (25.0, 600.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 15), (1.0, 8.0), linkedParams=[[0, 4, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 3, 16), (1000.0, 8000.0), linkedParams=[[0, 4,
0]]))
ffieldParams_.append(ForceFieldParam(
(0, 7, 8), (-1.0, 1.0))) #, linkedParams=[[0, 4, 0]]))
ffieldParams_.append(ForceFieldParam(
(0, 7, 6), (1000.0, 1000000.0))) # , linkedParams=[[0, 4, 0]]))
ffieldParams_.append(ForceFieldParam(
(0, 7, 7), (1.0, 1000.0))) # , linkedParams=[[0, 4, 0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
originalValues = [
ffield_.data[0][param.loc[1]][param.loc[2]] for param in ffieldParams_
]
print(originalValues)
lmpFile = LammpsFile("lammpsTest.lmp")
structToEnergyMap = KVR.readStructureEnergyValuesToDict(
"Si3N4_DATA/README.071820")
def ObjFunc(paramSetList):
return objectiveFunction_test2(paramSetList, ffieldParams=ffieldParams_, ffield=ffield_, \
lmpFile=lmpFile, structToEnergyMap=structToEnergyMap)
normBOpt = normalizationWrapper(paramRanges, ObjFunc, startingSamples=50)
normBOpt.learn(50)
bOpt = normBOpt.bOpt
print("best error prior to GD:")
print(max(normBOpt.bOpt.sampleResults.ravel()))
bestIndex = argmax(bOpt.sampleResults.ravel())
print("corresponding paramSet:")
print(
BoA.descaleSamples0to1(bOpt.sampleArray[bestIndex],
normBOpt.paramRanges))
x = bOpt.optimalParams()
x = BoA.descaleSamples0to1(x, normBOpt.paramRanges)
print("\n------------\n\nbest error after GD:")
print(ObjFunc(x))
print("corresponding paramSet:")
print(x)
print("\n------------\n")
print("The partial errors: ", objectiveFunction_test2(x,printPartials=True, ffieldParams=ffieldParams_, ffield=ffield_, \
lmpFile=lmpFile, structToEnergyMap=structToEnergyMap))
def test2():
# print("hello World",flush=True)
ffieldParams_ = []
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
print(ffield_.data[0][3][11])
#print(ffield_.data[0][3][12])
#print(ffield_.data[0][3][15])
#print(ffield_.data[0][3][16])
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (1.5, 6.0), linkedParams=[[0, 4, 0]])
]
#ffieldParams_.append(ForceFieldParam((0, 3, 12), (25, 600), linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 3, 15), (1, 8), linkedParams=[[0, 4, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 3, 16), (1000, 8000), linkedParams=[[0, 4, 0]]))
paramRanges = [val.paramRange for val in ffieldParams_]
lmpFile = LammpsFile("lammpsTest.lmp")
structToEnergyMap = {
"data.alpha_Si3N4_1.0": -303.46972,
"data.alpha_Si3N4_0.98": -301.88812,
"data.alpha_Si3N4_1.02": -301.83053,
"data.alpha_Si3N4_0.9": -264.85529,
"data.alpha_Si3N4_1.1": -274.56516
}
def ObjFunc(paramSetList):
return objectiveFunction_test2(paramSetList, ffieldParams=ffieldParams_, ffield=ffield_,\
lmpFile=lmpFile,structToEnergyMap=structToEnergyMap)
fig = plt.figure()
bOpt = BayesOptAlg(paramRanges, ObjFunc, startingSamples=2)
bGraph = fig.add_subplot(211)
acGraph = fig.add_subplot(212)
for i in range(10):
plotUtils.plotBayesOpt(bOpt, bGraph, acGraph)
bOpt.learn(1)
print("best error:", end="")
print(max(bOpt.sampleResults.ravel()))
plt.pause(0.5)
print(max(bOpt.sampleResults))
x = argmax(bOpt.sampleResults)
print(bOpt.sampleArray[x])
plotUtils.plotBayesOpt(bOpt, bGraph, acGraph)
plt.show()
def readParams(pathname):
'''
reads in the parameter file at pathname, and returns a split-line by line list (one line per
inner list, one word per inner list member)
:param pathname: String
the location of the param list
:return: 2d array/list of lists
the list containing the param files lines
'''
lines = fr.readFileAsLines(pathname)
lines = fr.removeWhiteSpaceAndComments(lines)
data = [line.split() for line in lines]
ffieldParams_ = []
for line in data:
ffieldParams_.append(ForceFieldParam((int(line[0])-1, int(line[1])-1, int(line[2])+2)\
, (float(line[4]), float(line[5])), linkedParams=[]))
temp = ffieldParams_[len(ffieldParams_)-1]
print(temp.loc)
return ffieldParams_
def test1():
# print("hello World",flush=True)
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff("Si3N4_DATA/ffield.original")
print(ffield_.data[0][3][11])
ffieldParams_ = [
ForceFieldParam((0, 3, 11), (0.5, 6.0), linkedParams=[[0, 4, 0]])
]
paramRanges = [val.paramRange for val in ffieldParams_]
def ObjFunc(paramList):
return objectiveFunction_test1(paramList,
ffieldParams=ffieldParams_,
ffield=ffield_)
bOpt = BayesOptAlg(paramRanges, ObjFunc)
bOpt.learn(10)
print(max(bOpt.sampleResults))
x = argmax(bOpt.sampleResults)
print(bOpt.sampleArray[x])
def geoFileReadingTest():
# print("hello World",flush=True)
ffieldParams_ = []
comm = MPI.COMM_WORLD
ffield_ = ForceField_Tersoff_OG("DATA2/Tersoff.BM.98.dat")
ffieldParams_ = [
ForceFieldParam((0, 4, 8), (0.01, 150.0), linkedParams=[[0, 0, 0]])
]
#ffieldParams_.append(ForceFieldParam((0, 4, 3), (1000.0, 100000.0), linkedParams=[[0, 0, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 4, 4), (2.0, 600.0), linkedParams=[[0, 0, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 4, 5), (-5.0, -0.01), linkedParams=[[0, 0, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 4, 6), (0.1, 100.0), linkedParams=[[0, 0, 0]]))
#ffieldParams_.append(ForceFieldParam((0, 4, 7), (0.0000001, 0.1), linkedParams=[[0, 0, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 4, 9), (25.0, 60000.0), linkedParams=[[0, 0, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 4, 12), (1.0, 8.0), linkedParams=[[0, 0, 0]]))
ffieldParams_.append(
ForceFieldParam((0, 4, 13), (100.0, 800000.0),
linkedParams=[[0, 0, 0]]))
# ffieldParams_.append(ForceFieldParam((0, 7, 8), (-1.0, 1.0))) # , linkedParams=[[0, 4, 0]]))
# ffieldParams_.append(ForceFieldParam((0, 7, 6), (1000.0, 1000000.0))) # , linkedParams=[[0, 4, 0]]))
# ffieldParams_.append(ForceFieldParam((0, 7, 7), (1.0, 1000.0))) # , linkedParams=[[0, 4, 0]]))'''
#ffieldParams_ = ParamReading.readParams("DATA2/params")
paramRanges = [val.paramRange for val in ffieldParams_]
originalValues = [
ffield_.data[0][param.loc[1]][param.loc[2]] for param in ffieldParams_
]
print(originalValues)
subprocess.call(["cp", "src/GeoFileReading/geo2data.exe", "DATA2"])
subprocess.call(["./DATA2/geo2data.exe", "DATA2/geo", "DATA2/ffield"])
os.remove("DATA2/geo2data.exe")
objList = readTrainsetFile("DATA2/EVTrainset.in", singleMeanSquaredError,
"")
def objWrapper(simpleParamList):
return -objectiveFunctionFromPartials(simpleParamList, ffieldParams_,
ffield_, objList)
print(objWrapper(originalValues))
normBOpt = normalizationWrapper(paramRanges,
objWrapper,
startingSamples=800)
normBOpt.learn(200)
bOpt = normBOpt.bOpt
print("best error prior to GD:")
print(max(normBOpt.bOpt.sampleResults.ravel()))
bestIndex = argmax(bOpt.sampleResults.ravel())
print("corresponding paramSet:")
print(
BoA.descaleSamples0to1(bOpt.sampleArray[bestIndex],
normBOpt.paramRanges))
x = bOpt.optimalParams()
x = BoA.descaleSamples0to1(x, normBOpt.paramRanges)
print("\n------------\n\nbest error after GD:")
print(objWrapper(x))
print("corresponding paramSet:")
print(x)