(C++) 在 HDF5 中编写数据库

Question

您好，这是我第一次使用 HDF5，我正在尝试使用手动创建的数据集创建数据库表。文件已创建并已部署，但是每次我尝试插入单个数据注册表（每次 cv::mat 为 1x6，因为我尝试插入整个数据集但我也无法插入）我收到以下消息：

H5Dwrite(): 文件选择+偏移不在范围内 专业：数据空间 次要：超出范围 在抛出 'H5::DataSetIException' 的实例后调用终止

我使用的代码如下

cv::Mat label;
int size = data.length()/n_features_objects;

size_t size_[2]={1,(hsize_t)n_features_objects};
hsize_t chunk[2]={1,(hsize_t)n_features_objects};
hsize_t max_size[2]={H5S_UNLIMITED,(hsize_t)n_features_objects};
H5::DSetCreatPropList prop;
prop.setChunk(2,chunk);
H5::DataSpace ds(2, size_,max_size);
H5::IntType datatype =  H5::PredType::NATIVE_UINT64;
objects = labelFile.createDataSet(objects_name.toStdString(),datatype, ds,prop);

//get size of the dataset
H5::DataSpace ds = objects.getSpace();
int dims = ds.getSimpleExtentNdims();
hsize_t rc [dims];
ds.getSimpleExtentDims(rc);
int rows = (int)rc[0];

for(int z=0; z<size;z++)
{
    hsize_t new_size[2]={1, (hsize_t)n_features_objects};
    labels.extend(new_size);

    cv::Mat label_oreg = cv::Mat::zeros(cv::Size(n_features_objects,1), CV_32SC1);
    //Adding data registries one by one
    label_oreg.at<int>(0,0) = label_oreg.at<int>(z,0);
    label_oreg.at<int>(0,1) = label_oreg.at<int>(z,1);
    label_oreg.at<int>(0,2) = label_oreg.at<int>(z,2);
    label_oreg.at<int>(0,3) = label_oreg.at<int>(z,3);
    label_oreg.at<int>(0,4) = label_oreg.at<int>(z,4);
    label_oreg.at<int>(0,5) = label_oreg.at<int>(z,5);

    hsize_t rows_ext[2] ={1, (hsize_t)n_features_objects};
    hsize_t offset[2] ={(hsize_t)1, 0};
    ds = labels.getSpace();
    ds.selectHyperslab(H5S_SELECT_SET,rows_ext, offset);
    H5::DataSpace mem_space(2, rows_ext);
    labels.write(label.data, H5::PredType::NATIVE_INT32, mem_space, ds);
}

有人知道我在这里缺少什么吗？我认为我的结果很奇怪，因为我总是在数据空间中注册了一个新的数据块，以便将其用作放置注册表。

Answer 1

根据您的问题的标题和描述，您似乎希望在不覆盖现有数据的情况下将数据添加（即追加）到 HDF5 数据集中，就像表（在关系数据库中）允许您添加行而不覆盖现有行。

如果这是正确的，这里有一个通用示例来说明如何在 C++ 中使用 HDFql 来完成（我不知道如何在其他 HDF5 C++ API 中做到这一点）：

// declare variable
unsigned long long values[6];

// create HDF5 file "my_file.h5" and use (i.e. open) it
HDFql::execute("CREATE AND USE FILE my_file.h5");

// create HDF5 dataset "my_dataset" of datatype unsigned long long (64 bit) of two dimensions (unlimited x 6)
HDFql::execute("CREATE CHUNKED(1, 6) DATASET my_dataset AS UNSIGNED BIGINT(0 TO UNLIMITED, 6)");

// register variable "values" for subsequent use (by HDFql)
HDFql::variableRegister(values);

// keep reading data (from hypothetical function "read") and store it in variable "values" until there is no more to read
while(read(values))
{
    // extend first dimension of dataset "my_dataset" one unit
    HDFql::execute("ALTER DIMENSION my_dataset TO +1");

    // insert (i.e. write) content of variable "values" into the last position of the first dimension of dataset "my_dataset" (thanks to a hyperslab selection)
    HDFql::execute("INSERT INTO my_dataset(-1:1:1:1) VALUES FROM MEMORY 0");
}

// unregister variable "values" as it is no longer used/needed (by HDFql)
HDFql::variableUnregister(values);