在过去的几个月里，我一直在进行一个非常广泛的研究项目。不幸的是，由于各种原因，我不能深入研究这个问题，或者大部分程序，但我遇到了一个我似乎无法理解的问题，所以除了尝试堆栈溢出之外，我已经没有其他选择了。
我可以展示的程序部分获取存储在2-D或3-D向量中的一些数据，并将它们传递给我正在构建的核岭回归函数。在这个函数中，我将各种数据集分成训练集和测试集，以便交叉验证。在找到内核并执行正则化之后，我将修改后的内核与测试数据一起传递到我构建的高斯-乔丹消除函数中，以解决Ax = b中的x。我遇到的问题是，在完成此函数后，程序崩溃并出现以下错误:
agnigen(28820,0x7fff7512b000) malloc: * error for object 0x7f9fa2a063f8:错误的校验和释放对象-对象可能在被释放后被修改。*在malloc_error_break中设置断点进行调试Abort trap: 6
现在我已经缩小了这个错误使用valgrind到x向量，它被传递给高斯-乔丹消去函数的引用。然而，我完全不知道为什么会发生这个错误，我将感谢你能给我的任何帮助。下面将是我可以显示的代码(因为它不放弃任何技术分类)，但如果你需要更多，我会尽我最大的努力提供它。谢谢你的帮助。
这是KRR函数被调用的地方:一个名为Run
的文件LearningAlgorithms::KRR(dataStorage,trainingFP,trainingForces,prediction);
这是KRR函数:
void LearningAlgorithms::KRR(DataStorage *dS, vector<vector<double> > trainingPoints, vector<vector<double> > trainingForces, vector<double> &prediction)
{
//variable declarations for KRR
int cvFolds = atoi(dS->getCV()[1].c_str());
int trainPartitionSize = (int) trainingPoints.size() / cvFolds;
double distance = 0.0;
vector<double> lambdas,sigmas,errors;
vector<vector<double> > alphas,trainingSet, testSet, randomizedTrainingPoints,trainingSetForces, testSetForces, randomizedTrainingForces, kernelMatrix;
vector<vector<vector<double > > > cvAlphas;
prediction.resize(trainingPoints.size());
trainingSet.resize(trainPartitionSize);
testSet.resize(trainingPoints.size() - trainPartitionSize);
trainingSetForces.resize(trainPartitionSize);
testSetForces.resize(trainingPoints.size() - trainPartitionSize);
for(int i = 0; i < trainingSet.size(); i++)
trainingSet[i].resize(trainingPoints[0].size());
for(int i = 0; i < testSet.size(); i++)
testSet[i].resize(trainingPoints[0].size());
for(int i = 0; i < trainingSetForces.size(); i++)
trainingSetForces[i].resize(1);
for(int i = 0; i < testSetForces.size(); i++)
testSetForces[i].resize(1);
//set sizes of hyper-parameter vectors based on whether cross validation was chosen or not
if(dS->getCV()[0] == "T" || dS->getCV()[0] == "t")
{
cout<<"Training model using "<<cvFolds<<"-fold Kernel Ridge Regression"<<endl;
lambdas.resize(cvFolds);
sigmas.resize(cvFolds);
errors.resize(cvFolds);
cvAlphas.resize(cvFolds);
for(int i = 0; i < lambdas.size(); i++)
{
lambdas[i] = Math::randomDouble(dS->getLambdaRange()[0],dS->getLambdaRange()[1]);
sigmas[i] = Math::randomDouble(dS->getSigmaRange()[0],dS->getSigmaRange()[1]);
}
}
else
{
cout<<"Training model using Kernel Ridge Regression without cross validation"<<endl;
lambdas.resize(1);
sigmas.resize(1);
errors.resize(1);
cvAlphas.resize(1);
lambdas[0] = Math::randomDouble(dS->getLambdaRange()[0],dS->getLambdaRange()[1]);
sigmas[0] = Math::randomDouble(dS->getSigmaRange()[0],dS->getSigmaRange()[1]);
}
//randomizes training points to diversify partitions during cross validation
Utilities::coordinatedRandomization2Darray(trainingPoints, randomizedTrainingPoints,trainingForces,randomizedTrainingForces);
//optimize lambda and sigma based on range provided
for(int lambda = 0; lambda < lambdas.size(); lambda++)
{
for(int sigma = 0; sigma < sigmas.size(); sigma++)
{
for(int fold = 0; fold < cvFolds; fold++)
{
int trainingIndex = 0;
int testIndex = 0;
int sum = 0;
//partitions training points into training and test sets based on cross validation
for(int training = 0; training < randomizedTrainingPoints.size(); training++)
{
sum = 0;
if(training >= trainPartitionSize * fold && training < trainPartitionSize * (fold + 1))
{
for(int i = 0; i < randomizedTrainingPoints[0].size(); i++)
trainingSet[trainingIndex][i] = randomizedTrainingPoints[training][i];
for(int i = 0; i < randomizedTrainingForces[0].size(); i++)
sum += randomizedTrainingForces[training][i];
trainingSetForces[trainingIndex][0] = sum;
trainingIndex++;
}
else
{
for(int i = 0; i < randomizedTrainingPoints[0].size(); i++)
testSet[testIndex][i] = randomizedTrainingPoints[training][i];
for(int i = 0; i < randomizedTrainingForces[0].size(); i++)
sum += randomizedTrainingForces[training][i];
testSetForces[trainingIndex][0] = sum;
testIndex++;
}
}
//clears the kernel matrix and resizes based on test set size
kernelMatrix.clear();
kernelMatrix.resize(testSet.size());
for(int km = 0; km < kernelMatrix.size(); km++)
kernelMatrix[km].resize(testSet.size());
//determines the kernel for a given test set
for(int testIndex = 0; testIndex < testSet.size(); testIndex++)
{
for(int ntrain = 0; ntrain < testSet.size(); ntrain++)
{
distance = 0.0;
for(int i = 0; i < testSet[testIndex].size(); i++)
distance += pow(testSet[testIndex][i] - testSet[ntrain][i],2);
kernelMatrix[testIndex][ntrain] = ... (sorry, can't show this)
}
}
//performs a lambda regularization on the kernel matrix to account for noise in the data
Math::regularization(kernelMatrix, lambdas[lambda]);
//solves linear system of equations to get our parameterization term (b) and our alpha vectors
//A = kernel matrix, x = alpha , b = forces : Ax = b
Math::gaussJordanElimination(kernelMatrix,alphas, testSetForces);
}
}
}
这是高斯-乔丹消去函数:
void Math::gaussJordanElimination(vector< vector< double > > A, vector<vector<double > > &x, vector< vector< double > > b)
{
int i, icolumn, irow, j, k, l, ll;
double ainvabs, dum, pivinv;
int n = A.size();
int m = A[0].size();
vector< int> indxc, indxr, ipivot;
vector< vector< double > > Ainv;
indxc.resize(n);
indxr.resize(n);
ipivot.resize(n);
x.resize(b.size());
//puts values from b into x, and A into A inverse
for(int i = 0; i < b.size(); i++)
{
x[i].resize(b[i].size());
for(int j = 0; j < x[i].size(); j++)
x[i][j] = b[i][j];
}
Ainv.resize(n);
for(int i = 0; i < A.size(); i++)
{
Ainv[i].resize(m);
for(int j = 0; j < Ainv[i].size(); j++)
Ainv[i][j] = A[i][j];
}
//sets default values for the pivot vector
for(j = 0; j < n; j++)
ipivot[j] = 0;
//main loop over the inverse dimensions for column reductions
for(i = 0; i < n; i++)
{
ainvabs = 0.0;
//search for pivot elements
for(j = 0; j < n; j++)
{
if(ipivot[j] != 1)
{
for(k = 0; k < n; k++)
{
if(ipivot[k] == 0)
{
if(fabs(Ainv[j][k]) >= ainvabs)
{
ainvabs = fabs(Ainv[j][k]);
irow = j;
icolumn = k;
}
}
}
}
}
//pivots determined, swap columns with rows (relabeling on columns/rows of matrix)
++(ipivot[icolumn]);
if(irow != icolumn)
{
for(l = 0; l < n; l++)
{
int temp1 = Ainv[irow][l];
int temp2 = Ainv[icolumn][l];
Ainv[irow][l] = temp2;
Ainv[icolumn][l] = temp1;
}
for(l = 0; l < m; l++)
{
int temp1 = x[irow][l];
int temp2 = x[icolumn][l];
x[irow][l] = temp2;
x[icolumn][l] = temp1;
}
}
indxr[i] = irow;
indxc[i] = icolumn;
if(Ainv[icolumn][icolumn] != 0.0)
pivinv = 1 / Ainv[icolumn][icolumn];
for(l = 0; l < n; l++)
Ainv[icolumn][l] *= pivinv;
for(l = 0; l < m; l++)
x[icolumn][l] *= pivinv;
for(ll = 0; ll < n; ll++)
{
if(ll !=icolumn)
{
dum = Ainv[ll][icolumn];
Ainv[ll][icolumn] = 0.0;
for(l = 0; l < n; l++)
Ainv[ll][l] -= Ainv[icolumn][l] * dum;
for(l = 0; l < m; l++)
x[ll][l] -= x[icolumn][l] * dum;
}
}
//reductions complete, interchange elements in reverse order as before
for(l = n - 1; l >= 0; l--)
{
if(indxr[l] != indxc[l])
{
for(k = 0; k < n; k++)
{
int temp1 = Ainv[k][indxr[l]];
int temp2 = Ainv[k][indxc[l]];
Ainv[k][indxr[l]] = temp2;
Ainv[k][indxc[l]] = temp1;
}
}
}
}
编辑
这些是我使用valgrind时得到的消息。他们非常混乱，请让我知道如果它会更容易缩短这些一点，我真的不知道哪些部分是有用的，哪些部分不是，所以我只是复制了整个消息。
==52421== Invalid write of size 8
==52421== at 0x100021B67:
Math::gaussJordanElimination(std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >&, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >) (type_traits:3599)
==52421== by 0x100025D3D: LearningAlgorithms::KRR(DataStorage*, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<double, std::__1::allocator<double> >&) (LearningAlgorithms.cpp:171)
==52421== by 0x100018FAE: Run::trainModel(DataStorage*) (Run.cpp:291)
==52421== by 0x100000E8A: main (Main.cpp:131)
==52421== Address 0x1021f62e0 is 240 bytes inside a block of size 312 free'd
==52421== at 0x10004A32D: free (vg_replace_malloc.c:534)
==52421== by 0x100018E91: Run::trainModel(DataStorage*) (new:176)
==52421== by 0x100000E8A: main (Main.cpp:131)
==52421== Block was alloc'd at
==52421== at 0x100049D81: malloc (vg_replace_malloc.c:303)
==52421== by 0x1009A27DD: operator new(unsigned long) (in /usr/lib/libc++.1.dylib)
==52421== by 0x100006AFF: _ZNSt3__16vectorINS0_IdNS_9allocatorIdEEEENS1_IS3_EEE18__construct_at_endIPS3_EENS_9enable_ifIXsr21__is_forward_iteratorIT_EE5valueEvE4typeES9_S9_m (memory:1596)
==52421== by 0x100006A05: std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >::vector(std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > > const&) (vector:1203)
==52421== by 0x100018D8D: Run::trainModel(DataStorage*) (vector:1195)
==52421== by 0x100000E8A: main (Main.cpp:131)
==52421== Invalid read of size 8
==52421== at 0x100021BA5: Math::gaussJordanElimination(std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >&, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >) (type_traits:3597)
==52421== by 0x100025D3D: LearningAlgorithms::KRR(DataStorage*, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<double, std::__1::allocator<double> >&) (LearningAlgorithms.cpp:171)
==52421== by 0x100018FAE: Run::trainModel(DataStorage*) (Run.cpp:291)
==52421== by 0x100000E8A: main (Main.cpp:131)
==52421== Address 0x102e54ec8 is 232 bytes inside a block of size 312 free'd
==52421== at 0x10004A32D: free (vg_replace_malloc.c:534)
==52421== by 0x100018E91: Run::trainModel(DataStorage*) (new:176)
==52421== by 0x100000E8A: main (Main.cpp:131)
==52421== Block was alloc'd at
==52421== at 0x100049D81: malloc (vg_replace_malloc.c:303)
==52421== by 0x1009A27DD: operator new(unsigned long) (in /usr/lib/libc++.1.dylib)
==52421== by 0x100006AFF: _ZNSt3__16vectorINS0_IdNS_9allocatorIdEEEENS1_IS3_EEE18__construct_at_endIPS3_EENS_9enable_ifIXsr21__is_forward_iteratorIT_EE5valueEvE4typeES9_S9_m (memory:1596)
==52421== by 0x100006A05: std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >::vector(std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > > const&) (vector:1203)
==52421== by 0x100018D8D: Run::trainModel(DataStorage*) (vector:1195)
==52421== by 0x100000E8A: main (Main.cpp:131)
valgrind: m_mallocfree.c:303 (SizeT get_bszB_as_is(Block *)): Assertion 'bszB_lo == bszB_hi' failed.
valgrind: Heap block lo/hi size mismatch: lo = 80, hi = 195611105019399384.
This is probably caused by your program erroneously writing past the
end of a heap block and corrupting heap metadata. If you fix any
invalid writes reported by Memcheck, this assertion failure will
probably go away. Please try that before reporting this as a bug.
host stacktrace:
==52421== at 0x2380501B3: ???
==52421== by 0x2380505CC: ???
==52421== by 0x2380505AA: ???
==52421== by 0x23805E6F8: ???
==52421== by 0x238040389: ???
==52421== by 0x23803F316: ???
==52421== by 0x238046D47: ???
==52421== by 0x23803E613: ???
==52421== by 0x238006E47: ???
==52421== by 0x7000012B7851: ???
==52421== by 0x700000DB1EEF: ???
==52421== by 0x700000DB1EEF: ???
==52421== by 0x700000009C9F: ???
==52421== by 0x700000009CAF: ???
sched status:
running_tid=1
Thread 1: status = VgTs_Runnable (lwpid 2567)
==52421== at 0x100021BA9: Math::gaussJordanElimination(std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >&, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >) (type_traits:3598)
==52421== by 0x100025D3D: LearningAlgorithms::KRR(DataStorage*, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<std::__1::vector<double, std::__1::allocator<double> >, std::__1::allocator<std::__1::vector<double, std::__1::allocator<double> > > >, std::__1::vector<double, std::__1::allocator<double> >&) (LearningAlgorithms.cpp:171)
==52421== by 0x100018FAE: Run::trainModel(DataStorage*) (Run.cpp:291)
==52421== by 0x100000E8A: main (Main.cpp:131)
所以经过5天的烦恼，我终于明白了。
在高斯-乔丹消去函数中:
int m = A[0].size()
int m = b[0].size()
as m决定了x的列数，如果x的大小与A相同，而b的维数与A不同，则矩阵乘法不起作用
这段代码中可能还有更多错误，但这个问题是导致程序崩溃的直接原因。