从C ++读取HDF5数据：如何读取此特定格式？

Question

I have an HDF5 file that I need to read in C++ but I'm having trouble as the format of the file seems a bit complicated...

The HDF5 file contains data saved from two devices. The data is a time series; it can be seen as two arrays, one for time and the second for the actual output from the device. The number of acquisitions is user-defined, but the number of acquisitions is the same for both devices (as their data is acquired at the same time).

For example, one file will contain the data from, let's say, 10 acquisitions, organized in something similar to:

/Device1/Acquisition_000
/Device1/Acquisition_001
[...]
/Device2/Acquisition_000
/Device2/Acquisition_001
[...]

Each acquisition will contain a time array and a data array.

Here's a screenshot of what HDFView sees in the file:

I though a "path" /Device2/Acquisition_000 was a dataset and tried to read it as such, but I'm having trouble. I then dumped the .h5 file using h5dump and got the following:

HDF5 "data.h5" {
GROUP "/" {
GROUP "Device1" {
    DATASET "Acquisition_000" {
        DATATYPE  H5T_COMPOUND {
            H5T_IEEE_F64BE "Time";
            H5T_IEEE_F64BE "Signal";
        }
        DATASPACE  SIMPLE { ( 270000 ) / ( 270000 ) }
        DATA {
        (0): {
            0,
            -0.0933597
            },
        (1): {
            2e-05,
            -0.0476648
            },
        (2): {
            4e-05,
            -0.0628964
            },
[...]

Now I don't know how I should read that structure. I saw the H5T_COMPOUND so I tried the compound example from http://www.hdfgroup.org/HDF5/doc/cpplus_RM/compound_8cpp-example.html but the dataset->read() does not seems to be able to read the data; valgrind reports accessing uninitialized data when std::cout'ing the data in a loop.

Another source of confusion is the "H5T_IEEE_F64BE" in the dump; isn't the BE part for big-endian? Both the machine generating the data and the one reading it are x86_64...

How can I read the "Time" and "Signal" arrays into C/C++ arrays?

For reference, here's my try at adapting the example:

const H5std_string FILE_NAME("data.h5");
const H5std_string DATASET_NAME("/Device1/Acquisition_000/");
H5File file(FILE_NAME, H5F_ACC_RDONLY);
DataSet dataset = file.openDataSet(DATASET_NAME);
const H5std_string MEMBER_TIME("time_name");
const H5std_string MEMBER_SIGN("signal_name");
// Try reading a single array:
CompType mtype3( sizeof(double) );
mtype3.insertMember(MEMBER_SIGN, 0, PredType::NATIVE_DOUBLE);
double *data_signal = new double[270000];
memset(data_signal, 0, 270000);
dataset.read(data_signal, mtype3);
// Print the data
for (int i = 0 ; i < 10 ; i++)
{
    std::cout << "data_signal[i=" << i << "] = " << data_signal[i] << std::endl;
}

and its output:

data_signal[i=0] = 0
data_signal[i=1] = 0
data_signal[i=2] = 0
data_signal[i=3] = 0
data_signal[i=4] = 0
data_signal[i=5] = 0
data_signal[i=6] = 0
data_signal[i=7] = 0
data_signal[i=8] = 0
data_signal[i=9] = 0

Additionally, Matlab can read the file using:

data = h5read('data.h5', '/Device1/Acquisition_000')
data = 

      Time: [270000x1 double]
    Signal: [270000x1 double]

Thanks a lot.

Answer 1

The member names are used to pull the correct data fields out of the file. "signal_name" doesn't match the name of the data in the file. Try using "Signal", as is visible from MATLAB and from the GUI viewer.

Eventually, you'll want to define a c++ structure that represents a time/signal pair, like the compound example:

struct dataPoint
{
    double timePoint;
    double signal;
};

CompType hdf5DataPointType( sizeof(dataPoint) );
hdf5DataPointType.insertMember(MEMBER_TIME, 0, PredType::NATIVE_DOUBLE);
hdf5DataPointType.insertMember(MEMBER_SIGN, sizeof(double), PredType::NATIVE_DOUBLE);

Then read directly into an array of dataPoint.