Skip to content

Commit bd6de6c

Browse files
shijiashuaishijiashuai
committed
remove some local pointers retured from inside a function
1 parent afc87c8 commit bd6de6c

File tree

2 files changed

+31
-36
lines changed

2 files changed

+31
-36
lines changed

‎mascot/svmModel.cu‎

Lines changed: 26 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -46,11 +46,12 @@ void svmModel::fit(const svmProblem &problem, const SVMParam &param) {
4646
probParam.C = 1.0;
4747
svmModel model;
4848
model.fit(subProblem, probParam);
49-
vector<float_point *> decValues;
49+
vector<vector<float_point> > decValues;
5050
//todo predict with cross validation
5151
model.predictValues(subProblem.v_vSamples, decValues);
5252
//binary model has only one sub-model
53-
sigmoidTrain(decValues.front(), subProblem.getNumOfSamples(), subProblem.v_nLabels, probA[k], probB[k]);
53+
sigmoidTrain(decValues.front().data(), subProblem.getNumOfSamples(), subProblem.v_nLabels, probA[k],
54+
probB[k]);
5455
probability = true;
5556
}
5657
svm_model binaryModel = trainBinarySVM(subProblem, param);
@@ -182,19 +183,20 @@ void svmModel::addBinaryModel(const svmProblem &problem, const svm_model &bModel
182183
}
183184

184185
void
185-
svmModel::predictValues(const vector<vector<float_point> > &v_vSamples, vector<float_point *> &decisionValues) const {
186+
svmModel::predictValues(const vector<vector<float_point> > &v_vSamples,
187+
vector<vector<float_point> > &decisionValues) const {
186188
decisionValues.clear();
187189
for (int k = 0; k < cnr2; ++k) {
188-
float_point *kernelValues = new float_point[v_vSamples.size() * supportVectors[k].size()];
190+
vector<float_point> kernelValues(v_vSamples.size() * supportVectors[k].size());
189191
computeKernelValuesOnFly(v_vSamples, supportVectors[k], kernelValues);
190-
decisionValues.push_back(
191-
predictLabels(kernelValues, (int) v_vSamples.size(), k));//TODO not return local pointer in function
192-
delete[] kernelValues;
192+
vector<float_point> decValues41BinaryModel(v_vSamples.size() * supportVectors[k].size());
193+
predictLabels(kernelValues, decValues41BinaryModel, k);
194+
decisionValues.push_back(decValues41BinaryModel);
193195
}
194196
}
195197

196198
vector<int> svmModel::predict(const vector<vector<float_point> > &v_vSamples, bool probability) const {
197-
vector<float_point *> decisionValues;
199+
vector<vector<float_point> > decisionValues;
198200
predictValues(v_vSamples, decisionValues);
199201
vector<int> labels;
200202
if (!probability) {
@@ -233,7 +235,7 @@ vector<int> svmModel::predict(const vector<vector<float_point> > &v_vSamples, bo
233235
return labels;
234236
}
235237

236-
float_point svmModel::sigmoid_predict(float_point decValue, float_point A, float_point B) const {
238+
float_point svmModel::sigmoidPredict(float_point decValue, float_point A, float_point B) const {
237239
double fApB = decValue * A + B;
238240
// 1-p used later; avoid catastrophic cancellation
239241
if (fApB >= 0)
@@ -242,7 +244,7 @@ float_point svmModel::sigmoid_predict(float_point decValue, float_point A, float
242244
return 1.0 / (1 + exp(fApB));
243245
}
244246

245-
void svmModel::multiclass_probability(const vector<vector<float_point> > &r, vector<float_point> &p) const {
247+
void svmModel::multiClassProbability(const vector<vector<float_point> > &r, vector<float_point> &p) const {
246248
int t, j;
247249
int iter = 0, max_iter = max(100, nrClass);
248250
double **Q = (double **) malloc(sizeof(double *) * nrClass);
@@ -301,7 +303,7 @@ void svmModel::multiclass_probability(const vector<vector<float_point> > &r, vec
301303

302304
vector<vector<float_point> > svmModel::predictProbability(const vector<vector<float_point> > &v_vSamples) const {
303305
vector<vector<float_point> > result;
304-
vector<float_point *> decValues;
306+
vector<vector<float_point> > decValues;
305307
predictValues(v_vSamples, decValues);
306308
for (int l = 0; l < v_vSamples.size(); ++l) {
307309
vector<vector<float_point> > r(nrClass, vector<float_point>(nrClass));
@@ -310,21 +312,20 @@ vector<vector<float_point> > svmModel::predictProbability(const vector<vector<fl
310312
for (int i = 0; i < nrClass; i++)
311313
for (int j = i + 1; j < nrClass; j++) {
312314
r[i][j] = min(
313-
max(sigmoid_predict(decValues[k][l], probA[k], probB[k]), min_prob), 1 - min_prob);
315+
max(sigmoidPredict(decValues[k][l], probA[k], probB[k]), min_prob), 1 - min_prob);
314316
r[j][i] = 1 - r[i][j];
315317
k++;
316318
}
317319
vector<float_point> p(nrClass);
318-
multiclass_probability(r, p);
320+
multiClassProbability(r, p);
319321
result.push_back(p);
320322
}
321323
return result;
322324
}
323325

324-
void
325-
svmModel::computeKernelValuesOnFly(const vector<vector<float_point> > &samples,
326-
const vector<vector<float_point> > &supportVectors,
327-
float_point *kernelValues) const {
326+
void svmModel::computeKernelValuesOnFly(const vector<vector<float_point> > &samples,
327+
const vector<vector<float_point> > &supportVectors,
328+
vector<float_point> &kernelValues) const {
328329
for (int i = 0; i < samples.size(); ++i) {
329330
for (int j = 0; j < supportVectors.size(); ++j) {
330331
//rbf kernel
@@ -338,7 +339,8 @@ svmModel::computeKernelValuesOnFly(const vector<vector<float_point> > &samples,
338339
}
339340
}
340341

341-
float_point *svmModel::predictLabels(const float_point *kernelValues, int nNumofTestSamples, int k) const {
342+
void svmModel::predictLabels(const vector<float_point> &kernelValues, vector<float_point> &classificationResult,
343+
int k) const {
342344
//get infomation from SVM model
343345
int nNumofSVs = (int) supportVectors[k].size();
344346
// int nNumofSVs = GetNumSV(pModel);
@@ -360,21 +362,21 @@ float_point *svmModel::predictLabels(const float_point *kernelValues, int nNumof
360362

361363
StorageManager *manager = StorageManager::getManager();
362364
long long int nMaxNumofFloatPoint = manager->GetFreeGPUMem();
363-
long long int nNumofPart = Ceil(nNumofSVs * nNumofTestSamples, nMaxNumofFloatPoint);
365+
long long int nNumofPart = Ceil(nNumofSVs * kernelValues.size(), nMaxNumofFloatPoint);
364366

365367
// cout << "cache size is: " << nMaxNumofFloatPoint << " v.s.. " << nNumofSVs * nNumofTestSamples << endl;
366368
// cout << "perform classification in " << nNumofPart << " time(s)" << endl;
367369

368370
//allocate memory for storing classification result
369-
float_point *pfClassificaitonResult = new float_point[nNumofTestSamples];
371+
// float_point *pfClassificaitonResult = new float_point[kernelValues.size()];
370372
//initialise the size of each part
371373
int *pSizeofPart = new int[nNumofPart];
372-
int nAverageSize = (int) (nNumofTestSamples / nNumofPart);
374+
int nAverageSize = (int) (kernelValues.size() / nNumofPart);
373375
for (int i = 0; i < nNumofPart; i++) {
374376
if (i != nNumofPart - 1) {
375377
pSizeofPart[i] = nAverageSize;
376378
} else {
377-
pSizeofPart[i] = nNumofTestSamples - nAverageSize * i;
379+
pSizeofPart[i] = kernelValues.size() - nAverageSize * i;
378380
}
379381
}
380382

@@ -388,7 +390,7 @@ float_point *svmModel::predictLabels(const float_point *kernelValues, int nNumof
388390
checkCudaErrors(cudaMemset(pfDevSVsYiAlpha, 0, sizeof(float_point) * nNumofSVs));
389391
// checkCudaErrors(cudaMemset(pnDevSVsLabel, 0, sizeof(int) * nNumofSVs));
390392

391-
checkCudaErrors(cudaMemcpy(pfDevSVYiAlphaHessian, kernelValues + i * nAverageSize * nNumofSVs,
393+
checkCudaErrors(cudaMemcpy(pfDevSVYiAlphaHessian, kernelValues.data() + i * nAverageSize * nNumofSVs,
392394
sizeof(float_point) * nNumofSVs * pSizeofPart[i], cudaMemcpyHostToDevice));
393395
checkCudaErrors(
394396
cudaMemcpy(pfDevSVsYiAlpha, pfSVsYiAlpha, sizeof(float_point) * nNumofSVs, cudaMemcpyHostToDevice));
@@ -403,12 +405,7 @@ float_point *svmModel::predictLabels(const float_point *kernelValues, int nNumof
403405

404406
//perform classification
405407
ComputeClassLabel(pSizeofPart[i], pfDevSVYiAlphaHessian,
406-
nNumofSVs, fBias, pfClassificaitonResult + i * nAverageSize);
407-
408-
if (pfClassificaitonResult == NULL) {
409-
cerr << "error in ComputeClassLabel" << endl;
410-
exit(-1);
411-
}
408+
nNumofSVs, fBias, classificationResult.data() + i * nAverageSize);
412409

413410

414411
//free memory
@@ -417,8 +414,6 @@ float_point *svmModel::predictLabels(const float_point *kernelValues, int nNumof
417414
checkCudaErrors(cudaFree(pfDevSVsYiAlpha));
418415
// checkCudaErrors(cudaFree(pnDevSVsLabel));
419416
}
420-
421-
return pfClassificaitonResult;
422417
}
423418

424419
/*

‎mascot/svmModel.h‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -27,20 +27,20 @@ class svmModel {
2727

2828
unsigned int inline getK(int i, int j) const;
2929

30-
float_point *predictLabels(const float_point *kernelValues, int, int) const;
30+
void predictLabels(const vector<float_point> &kernelValues, vector<float_point> &classificationResult, int) const;
3131

3232
float_point *ComputeClassLabel(int nNumofTestSamples,
3333
float_point *pfDevSVYiAlphaHessian, const int &nNumofSVs,
3434
float_point fBias, float_point *pfFinalResult) const;
3535

3636
void computeKernelValuesOnFly(const vector<vector<float_point> > &samples,
37-
const vector<vector<float_point> > &supportVectors, float_point *kernelValues) const;
37+
const vector<vector<float_point> > &supportVectors, vector<float_point> &kernelValues) const;
3838

3939
void addBinaryModel(const svmProblem &, const svm_model &, int i, int j);
4040

41-
float_point sigmoid_predict(float_point decValue, float_point A, float_point B) const;
41+
float_point sigmoidPredict(float_point decValue, float_point A, float_point B) const;
4242

43-
void multiclass_probability(const vector<vector<float_point> > &, vector<float_point> &) const;
43+
void multiClassProbability(const vector<vector<float_point> > &, vector<float_point> &) const;
4444

4545
void
4646
sigmoidTrain(const float_point *decValues, const int, const vector<int> &labels, float_point &A, float_point &B);
@@ -53,7 +53,7 @@ class svmModel {
5353

5454
vector<vector<float_point> > predictProbability(const vector<vector<float_point> > &) const;
5555

56-
void predictValues(const vector<vector<float_point> > &, vector<float_point *> &) const;
56+
void predictValues(const vector<vector<float_point> > &, vector<vector<float_point> > &) const;
5757
};
5858

5959

0 commit comments

Comments
 (0)