import multiprocessing as mp
pool=mp.pool(4)
def PrimitiveElasticGraphEmbedment(X, NodePositions, ElasticMatrix,
MaxNumberOfIterations=10, eps=0.01,
Mode = 1, FinalEnergy = "Base",
alpha = 0,
beta = 0,
prob = 1,
DisplayWarnings = True,
PointWeights=None, MaxBlockSize=100000000,
verbose=False, TrimmingRadius=float('inf'),
SquaredX=None):
N = X.shape[0]
if PointWeights is None:
PointWeights = np.ones((N, 1))
SpringLaplacianMatrix = ComputeSpringLaplacianMatrix(ElasticMatrix)
if SquaredX is None:
SquaredX = (X**2).sum(axis=1).reshape((N, 1))
partition, dists = PartitionData(X, NodePositions, MaxBlockSize,
SquaredX, TrimmingRadius)
if verbose or Mode == 2:
OldElasticEnergy, MSE, EP, RP = ComputePrimitiveGraphElasticEnergy(
NodePositions, ElasticMatrix, dists)
ElasticEnergy = 0
for i in range(MaxNumberOfIterations):
NewNodePositions = FitGraph2DataGivenPartition(
X, PointWeights, SpringLaplacianMatrix, partition)
if verbose or Mode == 2:
ElasticEnergy, MSE, EP, RP = ComputePrimitiveGraphElasticEnergy(
NewNodePositions, ElasticMatrix, dists)
if Mode == 1:
diff = ComputeRelativeChangeOfNodePositions(
NodePositions, NewNodePositions)
elif Mode == 2:
diff = (OldElasticEnergy - ElasticEnergy)/ElasticEnergy
if verbose:
print("Iteration ", (i+1), " difference of node position=", np.around(diff,5),
", Energy=", np.around(ElasticEnergy,5), ", MSE=", np.around(MSE,5), ", EP=", np.around(EP,5),
", RP=", np.around(RP,5))
if not np.isfinite(diff):
diff=0
if diff < eps:
break
elif i < MaxNumberOfIterations-1:
partition, dists = PartitionData(X, NewNodePositions, MaxBlockSize,
SquaredX, TrimmingRadius)
NodePositions = NewNodePositions
OldElasticEnergy = ElasticEnergy
if DisplayWarnings and not(diff < eps):
print("Maximum number of iterations (", MaxNumberOfIterations,
") has been reached. diff = ", diff)
if (FinalEnergy != "Base") or (not(verbose) and (Mode != 2)):
if FinalEnergy == "Base":
ElasticEnergy, MSE, EP, RP = ComputePrimitiveGraphElasticEnergy(
NewNodePositions, ElasticMatrix, dists)
elif FinalEnergy == "Penalized":
ElasticEnergy, MSE, EP, RP = ComputePenalizedPrimitiveGraphElasticEnergy(
NewNodePositions, ElasticMatrix, dists,alpha,beta)
EmbeddedNodePositions = NewNodePositions
return EmbeddedNodePositions, ElasticEnergy, partition, dists, MSE, EP, RP
def proxy(Dict):
return PrimitiveElasticGraphEmbedment(**Dict)
pool = pool)
def ApplyOptimalGraphGrammarOperation(X,
NodePositions,
ElasticMatrix,
opTypes,
AdjustVect = None,
SquaredX = None,
verbose = False,
MaxBlockSize = 100000000,
MaxNumberOfIterations = 100,
eps = .01,
TrimmingRadius = float('inf'),
Mode = 1,
FinalEnergy = "Base",
alpha = 1,
beta = 1,
EmbPointProb = 1,
AvoidSolitary = False,
AdjustElasticMatrix = None,
DisplayWarnings = True,
n_cores = 1,
MinParOp = 20,
multiproc_shared_variables = None,
pool = None):
NodePositionsArrayAll = []
ElasticMatricesAll = []
AdjustVectAll = []
if(n_cores > 1 and len(Valid_configurations) // (MinParOp + 1) > 1):
results=pool.map(proxy,[dict(X=X,
NodePositions = NodePositionsArrayAll[i],
ElasticMatrix = ElasticMatricesAll[i],
MaxNumberOfIterations = MaxNumberOfIterations,eps=eps,
Mode=Mode, FinalEnergy=FinalEnergy,
alpha=alpha,beta=beta,prob=EmbPointProb,
DisplayWarnings=DisplayWarnings,
PointWeights=None,
MaxBlockSize=MaxBlockSize,
verbose=False,
TrimmingRadius=TrimmingRadius, SquaredX=SquaredX) for i in Valid_configurations])
list_energies = [r[1] for r in results]
idx = list_energies.index(min(list_energies))
NewNodePositions, minEnergy, partition, Dist, MSE, EP, RP = results[idx]
AdjustVect = AdjustVectAll[idx]
NewElasticMatrix = ElasticMatricesAll[idx]
else:
minEnergy = np.inf
for i in Valid_configurations:
nodep, ElasticEnergy, part, dist, mse, ep, rp = PrimitiveElasticGraphEmbedment(X,
NodePositionsArrayAll[i],
ElasticMatricesAll[i], MaxNumberOfIterations, eps,
Mode=Mode, FinalEnergy=FinalEnergy,
alpha=alpha,beta=beta,prob=EmbPointProb,
DisplayWarnings=DisplayWarnings,
PointWeights=None,
MaxBlockSize=MaxBlockSize,
verbose=False,
TrimmingRadius=TrimmingRadius, SquaredX=SquaredX)
if(ElasticEnergy < minEnergy):
NewNodePositions = nodep
NewElasticMatrix = ElasticMatricesAll[i]
partition = part
AdjustVect = AdjustVectAll[i]
minEnergy = ElasticEnergy
MSE = mse
EP = ep
RP = rp
Dist = dist
return dict(NodePositions = NewNodePositions, ElasticMatrix = NewElasticMatrix,
ElasticEnergy = minEnergy, MSE = MSE, EP = EP, RP = RP, AdjustVect = AdjustVect, Dist = Dist)
if __name=="__main__":
while (UpdatedPG['NodePositions'].shape[0] < NumNodes) or GrammarOptimization:
UpdatedPG = ApplyOptimalGraphGrammarOperation(X,
UpdatedPG['NodePositions'],
UpdatedPG['ElasticMatrix'],
GrowGrammars[k],
MaxBlockSize = MaxBlockSize,
AdjustVect = UpdatedPG['AdjustVect'],
SquaredX = SquaredX,
verbose = False,
MaxNumberOfIterations = MaxNumberOfIterations,
eps = eps,
TrimmingRadius = TrimmingRadius,
Mode = Mode,
FinalEnergy = FinalEnergy,
alpha = alpha,
beta = beta,
EmbPointProb = EmbPointProb,
AvoidSolitary = AvoidSolitary,
AdjustElasticMatrix = AdjustElasticMatrix,
DisplayWarnings = DisplayWarnings,
n_cores=n_cores, MinParOp = MinParOp)
上面的代码适用于具有1000个数据点的小数据;当使用1000个以上的数据点时，它在Linux计算机中挂起而在Windows计算机中运行良好。我不知道如何进一步调试问题。想法/帮助进一步解决这个问题。我相信这段代码在Linux机器中与fork有关。任何想法都值得赞赏。
有关总代码链接https://github.com/j-bac/elpigraph-python/blob/master/elpigraph/src/grammar_operations.py