def runModel(self,x):
inputs = uq.stdVec()
for loopA in xrange(len(x)):
inputs.push_back(x[loopA])
outputs = uq.stdVec()
errorCode = uq.stdIntVec()
res = self.model.evalModelError(inputs,outputs,errorCode)
return res
def buildMultiresolutionMatrix(xTrain,maxOrder,maxDetailLevel):
# Construct samples
samples = uq.uqSamples(xTrain.shape[1])
currSample = uq.stdVec()
for loopA in range(xTrain.shape[0]):
currSample.clear()
for loopB in range(xTrain.shape[1]):
currSample.push_back(xTrain[loopA,loopB])
samples.addOneSample(currSample)
# Set Parameters for multiresolution basis
maxOrder = maxOrder
addLegendrePoly = True
addMW = True
useBinaryPartitions = False
mwMatType = uq.kMWFixedMaxDetailLevel
mwMatIncludeNullColumns = True
useExactMW = False
mwQuadOrder = 10
measure = npToStdMat(np.ones((2*mwQuadOrder,1)))
# print np.array(measure)
maxColumns = 0
maxDetailLevel = maxDetailLevel
# Construct Multiresolution Matrix on samples
mwMat = uq.uqMWMatrix(maxOrder,samples, \
addLegendrePoly,addMW,useBinaryPartitions, \
mwMatType,mwMatIncludeNullColumns, \
useExactMW,mwQuadOrder,measure, \
maxColumns,maxDetailLevel)
# Return Matrix
return mwMat
def npToStdMat(npMat):
res = uq.stdMat()
aux = uq.stdVec()
for loopA in range(npMat.shape[1]):
aux.clear()
for loopB in range(npMat.shape[0]):
aux.push_back(npMat[loopB,loopA])
res.push_back(aux)
return res
def setInitialParamGuess(self,useStartingParameterFromFile,startFromCentre,startParameterFile):
if(useStartingParameterFromFile):
# Set Boolean Variable
self.paramsFromFile = True
self.paramsFile = startParameterFile
# Read Parameters from file
self.x0 = np.loadtxt(startParameterFile)
elif(startFromCentre):
# Start from the average parameters
limits = uq.stdVec()
self.model.getParameterLimits(limits)
numParams = self.model.getParameterTotal()
self.x0 = np.zeros((numParams,))
for loopA in xrange(numParams):
self.x0[loopA] = 0.5*(limits[2*loopA + 0] + limits[2*loopA + 1])
else:
# Start from the default parameters
params = uq.stdVec()
self.model.getDefaultParams(params)
self.x0 = params
def buildRegressionMatrix(pTrainVals,ord):
# Construct samples
samples = uq.uqSamples(pTrainVals.shape[1])
currSample = uq.stdVec()
for loopA in range(pTrainVals.shape[0]):
currSample.clear()
for loopB in range(pTrainVals.shape[1]):
currSample.push_back(pTrainVals[loopA,loopB])
samples.addOneSample(currSample)
# Construct Polynomial Matrix on samples
polyMat = uq.uqPolyMatrix(samples,ord,uq.kPolyLegendre,uq.kMIPartialOrder)
# Return Matrix
return polyMat
def train(self,pMat,qTrainVals):
# Eval functional value with simple function
rhs = uq.stdVec()
for loopA in range(len(qTrainVals)):
rhs.push_back(qTrainVals[loopA])
# Construct the regression algorithm
bcs = uq.uqAlgorithmBCS();
# Set options: Print Progress to screen
bcs.opts.printProgressToScreen = True;
bcs.opts.printDBGMessages = False;
bcsCoeffs = uq.stdVec()
coeffPrec = uq.stdMat()
resNorm = 0.0
bcsReturn = bcs.run(pMat.getRowCount(),pMat.getColCount(),
rhs,pMat.getMatrix(),
bcsCoeffs,coeffPrec,resNorm);
# Print coefficients on screen
coeffs = np.array(bcsCoeffs)
# Return
return np.array(bcsCoeffs),np.array(coeffPrec),bcsReturn
# =============
# MAIN FUNCTION
# =============
if __name__ == "__main__":
# Construct samples
samples = uq.uqSamples()
samples.addVariable('Var1', uq.kSAMPLEUniform, -1.0, 1.0)
samples.addVariable('Var2', uq.kSAMPLEUniform, -1.0, 1.0)
samples.generateCartesianGrid(10, uq.kCC, uq.kHaarRange)
# Construct Polynomial Matrix on samples
polyMat = uq.uqPolyMatrix(samples, 5, uq.kPolyLegendre, uq.kMIPartialOrder)
# Eval functional value with simple function
rhs = uq.stdVec()
rhs.resize(polyMat.getRowCount())
for loopA in xrange(polyMat.getRowCount()):
rhs[loopA] = samples.getValuesAt(loopA, 0) + samples.getValuesAt(
loopA, 1)
# Construct the regression algorithm
bcs = uq.uqAlgorithmBCS()
# Set options: Print Progress to screen
bcs.opts.printProgressToScreen = False
bcs.opts.printDBGMessages = False
bcsCoeffs = uq.stdVec()
coeffPrec = uq.stdMat()
resNorm = 0.0
polyIntLegendre = uq.uqPolyBasis(uq.kPolyLegendre, currOrder + 1)
polyIntHermite = uq.uqPolyBasis(uq.kPolyHermite, currOrder + 1)
currentLoc = 0.0
for loopA in xrange(steps):
currentLoc = loopA / float(steps - 1)
results[loopA, 0] = polyIntMonomial.evaluate(currentLoc, currOrder)
results[loopA, 1] = polyIntLegendre.evaluate(currentLoc, currOrder)
results[loopA, 2] = polyIntHermite.evaluate(currentLoc, currOrder)
# ORTHOGONAL POLYNOMIALS
# Define the number of quadrature points
quadLevel = 10
# Define the Uniform Measure
measureAtQuadPoints = uq.stdVec(2 * quadLevel, 0.0)
for loopA in xrange(2 * quadLevel):
measureAtQuadPoints[loopA] = 1.0
orthoPoly = uq.uqOrthoPolyBasis(currOrder + 1, quadLevel,
measureAtQuadPoints)
currentLoc = 0.0
for loopA in xrange(steps):
currentLoc = loopA / float(steps - 1)
results[loopA, 3] = orthoPoly.evaluate(currentLoc, currOrder)
# MULTIWAVELETS
mwInt = uq.uqMWBasis(currOrder + 1, quadLevel)
currentLoc = 0.0
for loopA in xrange(steps):
# EVALUATE RANGE
#tmpLimits = uq.stdVec()
#approx.getExtremes(tmpLimits)
#measureSize = fabs(tmpLimits[1] - tmpLimits[0])
#print 'CURRENT LIMITS: %f %f' % (tmpLimits[0],tmpLimits[1])
# FORM SAMPLES USING INTEGRATION GRID IN 1D
mwOrder = 2
mwQuadOrder = 30
measure1DGridPoints = uq.uqSamples()
measure1DGridPoints.addVariable('grid1D', uq.kSAMPLEUniform, 0.0, 1.0)
measure1DGridPoints.generateCartesianGrid(mwQuadOrder, uq.kDoubleCC,
uq.kHaarRange)
intLocations = uq.stdVec()
#scale1DGridOnPartition(measure1DGridPoints,tmpLimits[0],tmpLimits[1],intLocations);
scale1DGridOnPartition(measure1DGridPoints, 0.0, 1.0, intLocations)
# EVALUATE MARGINAL AT QUADRATURE POINTS
measure = uq.stdMat()
tmp = uq.stdVec()
currInt = 0.0
currLoc = 0.0
currVal = 0.0
for loopA in xrange(intLocations.size()):
currLoc = intLocations[loopA]
#currVal = approx.evaluate(currLoc);
currVal = 1.0
currInt += currVal * measure1DGridPoints.getWeightAt(
loopA, measure1DGridPoints.getMaxWeightOrder())
# =============
if __name__ == "__main__":
# Assign Dataset
data = da.daData_Scalar_MultiplePatients()
data.readFromFile('tutorial.csv')
# Construct Specific Model
myModel = cm.cmTutorial()
# Assign Dataset
currentColumn = 1
myModel.setData(data, currentColumn)
# Get Default Input Parameter Values
inputs = uq.stdVec()
myModel.getDefaultParams(inputs)
# Solve Model
outputs = uq.stdVec()
errorCodes = uq.stdIntVec()
ll = myModel.evalModelError(inputs, outputs, errorCodes)
print 'Model Results'
print ''
print 'Final Location: %f' % (outputs[0])
print 'Total Time: %f' % (outputs[1])
print 'Maximum Height: %f' % (outputs[2])
print ''
print 'Resulting Log-likelihod: %f' % (ll)
def main(fileName, power, comm):
# MPI Init
rank = comm.Get_rank()
size = comm.Get_size()
# Set Model
model = cm.cmBertiniSolverModel(fileName)
modelType = fileName.split("_")[1]
# SEt power
model.setExponent(1 / power)
# add solution
sol = uq.stdVec()
# for loopA in range(9):
# sol.push_back(0.0)
# model.addSolution(sol) # Set DREAM Parameters
totChains = size
totGenerations = 30000
totalCR = 3
totCrossoverPairs = 5
dreamGRThreshold = 1.2
dreamJumpStep = 10
dreamGRPrintStep = 10
# Set OUTPUT Files
dreamChainFileName = 'chain_GR_000000.txt'
dreamGRFileName = 'gr_GR.txt'
# Set Restart File
# No restart Simulation
dreamRestartReadFileName = ''
# string dreamRestartReadFileName = "restart_read_GR.txt";
# Write restart file just in case
dreamRestartWriteFileName = 'restart_write_GR.txt'
# Set Prior Information
usePriorFromFile = False
priorFileName = ''
priorModelType = 0
# Initialize DREAM Action
dream = ac.acActionDREAM(totChains, totGenerations, totalCR,
totCrossoverPairs, dreamChainFileName,
dreamGRFileName, dreamGRThreshold, dreamJumpStep,
dreamGRPrintStep, dreamRestartReadFileName,
dreamRestartWriteFileName, usePriorFromFile,
priorFileName, priorModelType)
# Set Model
dream.setModel(model)
# Run MCMC Simulation
dream.go()
# Add Barrier
comm.Barrier()
# Post Process the Results
if (rank == 0):
debugMode = False
burnInPercent = 0.1
dream.postProcess(debugMode, burnInPercent)
# Rename File
shutil.move('paramTraces.txt',
'Datafiles/paramTraces_' + str(modelType))
