h5py：如何重命名尺寸？ [英] h5py : how to rename dimensions?

问题描述

我创建了一个句柄为fw的新文件。

I created a new file whose handle is fw.

fw.create_dataset('grp1/varname',data=arr)

在此命令之前创建了组。 arr是一个尺寸为（2,3）的numpy数组。文件创建成功。但是，这些维度分别命名为phony_0和phony_1。

The groups are created before this command. arr is a numpy array with dimensions (2,3). The file is created successfully. However, the dimensions are named phony_0, and phony_1. How do I change them to say m and n ?

一般来说，如何在组内创建维，然后将变量与它们关联？

In general how does one create dimensions within a group and then associate variables with them?

我尝试过，

fw['grp1/varname'].dims[0].label = 'm'

但这并没有达到预期的效果。

But this does not have the desired effect.

ncdump -h 在创建的文件上显示：

ncdump -h on the created file shows :

group: grp1 {

        dimensions:
                phony_dim_0 = 2 ;
                phony_dim_1 = 3 ;

        variables:

                float varname(phony_dim_0, phony_dim_1) ;
                        string varname:DIMENSION_LABELS = "m", NIL, NIL ;
        } // group grp1

谢谢

print（[fw ['grp1 / varname']。dims]中dim的dim.label
确实产生一致的输出。
[u’m，u’]

print([ dim.label for dim in fw['grp1/varname'].dims]) does produce consistent output. [u'm', u'']

似乎hdffile没有提供将维与组关联的规定。但是，varname是变量。如何获得：

It does seem that hdffiles do not have provision to associate dimensions with groups. However varname is a variable. How does one get :

   variables:
            float varname(m, phony_dim_1) ;
                    string varname:DIMENSION_LABELS = "m", NIL ;
    } // group grp1

吗？我确实使用h5dump尝试了其他选项。

in the output of ncdump -h or h5dump ? I did try different options with h5dump.

谢谢。

推荐答案

部分问题可能是您使用 ncdump 。

Part of the problem may your use of ncdump.

我可以制作一个简单的文件，并设置<$数据集的c $ c> dims 标签：

I can make a simple file, and set the dims label for a dataset:

In [420]: import h5py
In [421]: f = h5py.File('testdim.h5','w')
In [422]: ds = f.create_dataset('grp1/varname', data = np.arange(10))
In [423]: ds
Out[423]: <HDF5 dataset "varname": shape (10,), type "<i8">

查看 dims 属性：

In [424]: ds.dims
Out[424]: <Dimensions of HDF5 object at 140382697336904>
In [426]: ds.dims[0]
Out[426]: <"" dimension 0 of HDF5 dataset at 140382697336904>
In [427]: ds.dims[0].label
Out[427]: ''
In [428]: ds.dims[0].label = 'm'

In [436]: dd=ds.dims[0]
In [437]: dd?
Type:        DimensionProxy
String form: <"m" dimension 0 of HDF5 dataset at 140382697336904>
Length:      0
File:        ~/.local/lib/python3.6/site-packages/h5py/_hl/dims.py
Docstring:   Represents an HDF5 "dimension".
In [439]: dd.values()
Out[439]: []
In [440]: dd.label
Out[440]: 'm'

该组没有点心：

In [442]: g = f['grp1']
In [443]: g
Out[443]: <HDF5 group "/grp1" (1 members)>
In [444]: g.dims
AttributeError: 'Group' object has no attribute 'dims'

In [446]: f.flush()

具有 h5dump ：

1902:~/mypy$ h5dump testdim.h5 
HDF5 "testdim.h5" {
GROUP "/" {
   GROUP "grp1" {
      DATASET "varname" {
         DATATYPE  H5T_STD_I64LE
         DATASPACE  SIMPLE { ( 10 ) / ( 10 ) }
         DATA {
         (0): 0, 1, 2, 3, 4, 5, 6, 7, 8, 9
         }
         ATTRIBUTE "DIMENSION_LABELS" {
            DATATYPE  H5T_STRING {
               STRSIZE H5T_VARIABLE;
               STRPAD H5T_STR_NULLTERM;
               CSET H5T_CSET_ASCII;
               CTYPE H5T_C_S1;
            }
            DATASPACE  SIMPLE { ( 1 ) / ( 1 ) }
            DATA {
            (0): "m"
            }
         }
      }
   }
}
}

其中 ncdump 旨在显示 netcdf 文件：

1902:~/mypy$ ncdump -h testdim.h5 
netcdf testdim {

group: grp1 {
  dimensions:
    phony_dim_0 = 10 ;
  variables:
    int64 varname(phony_dim_0) ;
        string varname:DIMENSION_LABELS = "m" ;
  } // group grp1
}

c $ c> h5df 格式没有组尺寸； ncdump 为此创建了一个虚拟属性。

As best I can the h5df format does not have group dimensions; ncdump creates a dummy attribute for that.

要重申您先前问题的答案，请参阅HDF5尺寸文档是：

To reiterate the answer to your previous question, the documentation of HDF5 dimensions is:

http ：//docs.h5py.org/en/latest/high/dims.html

https://www.unidata.ucar.edu/software/netcdf/docs/interoperability_hdf5.html

对于HDF5文件

如果未使用尺寸比例，则netCDF-4仍然可以编辑

If dimension scales are not used, then netCDF-4 can still edit the file, and will invent anonymous dimensions for each variable shape.

NETCDF具有共享尺寸，HDF5具有尺寸比例。它们并不完全相同。

NETCDF has shared dimensions, HDF5 has dimension scales. They aren't quite the same.

http://www.stcorp.nl/beat/documentation/harp/conventions/hdf5.html

在HDF5数据模型中，没有共享尺寸的概念（与netCDF不同）。 HDF5数据集的形状被指定为尺寸长度的列表。但是，netCDF-4库使用HDF5作为其存储后端。它使用HDF5尺寸比例尺表示共享尺寸。

In the HDF5 data model there is no concept of shared dimensions (unlike netCDF). The shape of an HDF5 dataset is specified as a list of dimension lengths. However, the netCDF-4 library uses HDF5 as its storage backend. It represents shared dimensions using HDF5 dimension scales.

这篇关于h5py：如何重命名尺寸？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！