eipoa's profile - activity

        Ptr<SVM> svm = SVM::create();
        svm->setType(SVM::C_SVC);
        svm->setKernel(3);
        svm->setDegree(3); // have to set the param, runtime error
        svm->setTermCriteria(TermCriteria(TermCriteria::MAX_ITER + TermCriteria::EPS, 1000, 0.01));
        svm->trainAuto(_Samples, 10, SVM::getDefaultGrid(SVM::C), SVM::getDefaultGrid(SVM::GAMMA),
        SVM::getDefaultGrid(SVM::P), SVM::getDefaultGrid(SVM::NU), SVM::getDefaultGrid(SVM::COEF),
        SVM::getDefaultGrid(SVM::DEGREE), true);
        svm->predict(Test, results);

using LIBSVM with sigmoid can achieve 86% of accuracy, but using cv can only get about 40%. what's wrong with my code?

the performance of sigmoid kernel in opencv is quite worst than it in LIBSVM. but other kernels look same as them in LIBSVM

I want to test which combinations of features are more important. the varidx works when using bayes classifier. i thing they shuld be the same cause they are all the children of StatModel. anyway, i think that maybe after useing setTrainTestSplitRatio() to divied data into train and test, the varidx may work well. but unfortunately, after this it doesn't work when i call pridect(). i mean the following the steps donot work at all. 1. setTrainTestSplitRatio 2. train 3. predict i forgot the error message, it seems that data type error., but after spliting, no way to change data type. i donot know whether the ml pass the test.

OpenCV Error: Assertion failed (samples.cols == var_count && samples.type() == CV_32F) in predict, file C:/repo/mingw-w64-opencv/src/opencv-3.2.0/modules/ml/src/svm.cpp, line 1930

vector<int> Idx {1,2,3,4,5,6,7,8,9,10,11,13};//0,1,2,3,4,5,6,7,8,9,10,11,12,13; (14 features)
Mat varIdx(Idx);
Ptr<TrainData> _Samples = TrainData::create(Train, ROW_SAMPLE, TrainResp, _Params.getVarIdx());
Ptr<NormalBayesClassifier> normalBayesClassifier = StatModel::train<NormalBayesClassifier>(_Samples);
normalBayesClassifier->predict(Test, results);
......
svm->trainAuto(_Samples);
svm->predict(Test, results);// error

the varidx for svm doesnot work, but for Bayes it works, at least the program doesnot crash. I checked some answers from google, someone said varidx should be like this {0,1,1,1,1,0,1...}, it is so strange, i cannot believe it.