我试图在n维数据上运行kmeans算法。
我有N分,每个分数都有{ x, y, z, ... , n }
个功能。
我的代码如下:
cv::Mat points(N, n, CV_32F);
// fill the data points
cv::Mat labels; cv::Mat centers;
cv::kmeans(points, k, labels, cv::TermCriteria(CV_TERMCRIT_ITER|CV_TERMCRIT_EPS, 1000, 0.001), 10, cv::KMEANS_PP_CENTERS, centers);
问题是kmeans算法遇到了分段错误。
感谢任何帮助
更新
Miki和Micka如何说上面的代码是正确的!
我在"填写数据点"所以我破坏了记忆
答案 0 :(得分:1)
代码看起来不错。您必须选择每列1维的数据。
你能尝试运行这个例子吗?
// k-means
int main(int argc, char* argv[])
{
cv::Mat projectedPointsImage = cv::Mat(512, 512, CV_8UC3, cv::Scalar::all(255));
int nReferenceCluster = 10;
int nSamplesPerCluster = 100;
int N = nReferenceCluster*nSamplesPerCluster; // number of samples
int n = 10; // dimensionality of data
// fill the data points
// create n artificial clusters and randomly seed 100 points around them
cv::Mat referenceCenters(nReferenceCluster, n, CV_32FC1);
//std::cout << referenceCenters << std::endl;
cv::randu(referenceCenters, cv::Scalar::all(0), cv::Scalar::all(512));
//std::cout << "FILLED:" << "\n" << referenceCenters << std::endl;
cv::Mat points = cv::Mat::zeros(N, n, CV_32FC1);
cv::randu(points, cv::Scalar::all(-20), cv::Scalar::all(20)); // seed points around the center
for (int j = 0; j < nReferenceCluster; ++j)
{
cv::Scalar clusterColor = cv::Scalar(rand() % 255, rand() % 255, rand() % 255);
//cv::Mat & clusterCenter = referenceCenters.row(j);
for (int i = 0; i < nSamplesPerCluster; ++i)
{
// creating a sample randomly around the artificial cluster:
int index = j*nSamplesPerCluster + i;
//samplesRow += clusterCenter;
for (int k = 0; k < points.cols; ++k)
{
points.at<float>(index, k) += referenceCenters.at<float>(j, k);
}
// projecting the 10 dimensional clusters to 2 dimensions:
cv::circle(projectedPointsImage, cv::Point(points.at<float>(index, 0), points.at<float>(index, 1)), 2, clusterColor, -1);
}
}
cv::Mat labels; cv::Mat centers;
int k = 10; // searched clusters in k-means
cv::kmeans(points, k, labels, cv::TermCriteria(CV_TERMCRIT_ITER | CV_TERMCRIT_EPS, 1000, 0.001), 10, cv::KMEANS_PP_CENTERS, centers);
for (int j = 0; j < centers.rows; ++j)
{
std::cout << centers.row(j) << std::endl;
cv::circle(projectedPointsImage, cv::Point(centers.at<float>(j, 0), centers.at<float>(j, 1)), 30, cv::Scalar::all(0), 2);
}
cv::imshow("projected points", projectedPointsImage);
cv::imwrite("C:/StackOverflow/Output/KMeans.png", projectedPointsImage);
cv::waitKey(0);
return 0;
}
我在那里围绕人工聚类中心创建10维数据。为了显示我将它们投影到2D,得到这个结果: