HDF5 (.h5, .hdf5)

Background & Context

    • MIME type: application/x-hdf5
    • HDF data format Version 5.
    • General-purpose format for representing multidimensional datasets and images.
    • Datasets with compound data structures are supported.
    • Used for storage, management, and exchange of scientific data.
    • HDF is an acronym for Hierarchical Data Format.
    • Originally developed by the US National Center for Supercomputing Applications (NCSA).
    • Currently maintained by The HDF Group.
    • Binary file format.
    • Incompatible with HDF Version 4 and earlier.

Import & Export

  • Import["file.h5"] imports an HDF5 file, returning the names of the datasets stored in the file.
  • Import["file.h5",elem] imports the specified element from an HDF5 file.
  • The import format can be specified with Import["file","HDF5"] or Import["file",{"HDF5",elem,}].
  • Export["file.h5",expr] exports a numeric array to HDF5.
  • Export["file.h5",{expr1,},{"Datasets", {"dataset1",}}] creates an HDF5 file, storing the data arrays {expr1,} as separate datasets.
  • Export["file.h5",expr,elem] exports the specified element to an HDF5 file.
  • Export["file.h5",elem1->expr1,,"Rules"] uses rules to specify the elements to be exported.
  • See the following reference pages for full general information:
  • Import, Exportimport from or export to a file
    CloudImport, CloudExportimport from or export to a cloud object
    ImportString, ExportStringimport from or export to a string
    ImportByteArray, ExportByteArrayimport from or export to a byte array

Import Elements

  • General Import elements:
  • "Elements" list of elements and options available in this file
    "Summary"summary of the file
    "Rules"list of rules for all available elements
  • Structure elements:
  • "Datasets"names of all datasets
    "Groups"names of all groups
    "StructureGraph"a directed graph showing the structure of the datasets
    {"StructureGraph, groupname}a graph showing the structure under groupname
    "StructureGraphLegend"legend for the structure graph
    "Summary"summary of properties
  • Names of groups and datasets are given as the absolute paths starting with the root group name "/".
  • Import by default uses the "Datasets" element for the HDF5 format.
  • Data representation elements:
  • "Data"all datasets imported as an association
    {"Data",n} or nn^(th) dataset
    {"Data",dataset} or datasetnamed dataset
    {"Data",groupname}an association of all datasets under groupname
    {"Data",groupname,lev}an association of datasets under groupname up to level lev
  • The following basic data types are supported:
  • "Integer8"8-bit integers
    "Integer16"16-bit integers
    "Integer32"32-bit integers
    "Integer64"64-bit integers
    "UnsignedInteger8"8-bit unsigned integers
    "UnsignedInteger16"16-bit unsigned integers
    "UnsignedInteger32"32-bit unsigned integers
    "UnsignedInteger64"64-bit unsigned integers
    "Real32"IEEE singleprecision numbers
    "Real64"IEEE doubleprecision numbers
    "String"string of ASCII characters
  • The following structured data types are supported:
  • "ByteArray"a ByteArray of an arbitrary length
    "Array"an array of any supported data format
    "Enum"an enumeration
    "Compound"a compound dataset consisting of any other data format and other compound datasets
  • Complex numbers are typically stored and imported as compound types.
  • Metadata elements:
  • "Attributes"attributes of all groups and datasets
    "DataEncoding"specifies how each dataset is compressed
    "DataFormat"type used to represent each dataset
    "Dimensions"data dimensions of each dataset
    {"metadata",n}metadata of the n^(th) dataset
    {"metadata",dataset}metadata of the named dataset
  • The following data encodings are supported:
  • Noneno data compression is used
    "Fletcher32"adds the Fletcher checksum
    "GZIP"GZIP compression
    "ScaleOffset"performs a scale and/or offset operation
    "Shuffle"reorders so that consistent byte positions are placed together
    "SZIP"SZIP compression (Import only)
  • A single dataset can have multiple encodings, which will be specified as a list {enc1,enc2,}.

Export Elements

  • General Export element:
  • "Rules"a list of imported elements in the form of elemexpr
  • Export["file.h5",{elem1->expr1,},"Rules"] uses rules to specify the elements to be exported.
  • Available Export elements:
  • "Attributes"attributes associated to any object
    "Datasets"datasets and their associated elements
    "Groups"group names and their associated elements
    "NamedDataTypes"named data types and their associated elements
  • With the "Attributes" element, the following expressions can be given:
  • attrattributes associated to the root group "/"
    {path1attr1,}attributes attri associated to the specified pathi
  • Attributes attri should be given in the form "attname"->attval.
  • With the "Groups" element, the following expressions can be given:
  • {"gr1","gr2",}a list of group paths
    {"gr1"grdata1,}list of groups and their associated data
  • Group data grdatai can take the following keys:
  • "Attributes"group attributes
    "HardLinks"hard links to other objects
    "SoftLinks"soft links to other objects
  • Soft and hard links can be specified using "linkname"path.
  • With the "Datasets" element, the following expressions can be given:
  • datastores data under "Dataset1"
    {"name1"data1,}a list of dataset names and their associated data
    {"name1"ds1,}specifies each dataset dsi using a list of rules
  • Datasets dsi can take the following keys:
  • "Attributes"dataset attributes
    "Data"array of data
    "DataFormat"data type
    "MaxDimensions"list of maximal dimensions for the dataset
    "SpaceSelection"part of the dataspace where the data will be written
  • With the "NamedDataTypes" element, the following expressions can be given:
  • {"name1"type1,}a data type
    {"name1"<|"Type"type1,"Attributes"att1|>,}an association specifying a type and its attributes
  • The type specification typei can take the following forms:
  • "simpletype"a simple data type such as "Integer64"
    <|"Class""ByteArray",|>takes "Length" and "Tag" keys
    <|"Class""Array",|>takes "DataFormat" and "Dimensions" keys
    <|"Class""Compound",|>takes a "Structure" key


  • Import and Export options:
  • "ComplexKeys"Automatickeys for complex interpretation and export
  • By default, {"Re","Im"} are used as complex keys. Other settings include:
  • Noneno complex interpretation (import only)
    Automaticuse automatic keys
    {key1,key2}use explicit keys
  • Import option:
  • "TakeElements"Allsubset of elements to import
  • "TakeElements" can take the following values:
  • {elem1,elem2,}list of elements elemi
    {m;;n;;s,...}elements m through n in steps of s
    {opt1val1,}list of suboptions
  • The following suboptions opti are available for taking elements:
  • "Offset"{0,0,}the offset along the dimensions of the dataset
    "Count"Allthe number of block to be imported along each dimension
    "Stride"{1,1,}the step between beginnings of the blocks
    "Block"{1,1,}number of elements in each block
  • Export option:
  • ByteOrdering$ByteOrderingwhat byte ordering to use
    OverwriteTargetTruewhether to overwrite an existing file
    "AppendMode""Extend"how to append to existing objects
  • Using OverwriteTarget->"Append", new objects may be added to an existing file.
  • Possible settings for "AppendMode" include:
  • "Extend"extends existing objects, if possible (default)
    "Overwrite"overwrites exisiting objects
    "Preserve"preserves existing objects


open allclose all

Basic Examples  (1)

Show the datasets stored in a sample file:

Get the file summary:

Show the structure of the file:

Scope  (12)

Import  (7)

Show all elements available in the file:

Import, by default, returns the list of datasets in the file:

Import the file structure, specifying the format explicitly:

Import contents of a dataset by specifying its name:

Import dimensions and data format for all datasets in the file:

Import dimensions and data format of a single dataset:

Import 8-bit RGB raster data and render it as an Image object:

Export  (5)

Export a matrix to HDF5:

Show the datasets contained in this file:

Export a named dataset:

Export two matrices to HDF5:

Export a named dataset with given data type:

Import the data format:

Export a named data type and a group that links to it:

Import the file structure:

Import Elements  (22)

Attributes  (4)

Import attributes of all objects in the file:

Import attributes of a specific dataset:

Import attributes of the second dataset in the file:

The order of datasets in the file can be checked by calling:

Import attributes of multiple objects:

Data  (4)

Get data from all datasets in the file:

Import data from the "Complex64" dataset inside the "Complex" group:

Import data from the third dataset in the file:

Get data from all datasets in a given group:

Import data from every dataset or group in a list:

DataEncoding  (1)

Check what filters were applied to each dataset in the file:

DataFormat  (2)

Get data type description for every dataset in the file:

Simple numeric and string types have one-word descriptions:

"DataFormat" for a compound type shows class and structure:

"DataFormat" for enumerated types includes class, base data format and a list of values and names:

"DataFormat" for array types includes class, base data format and dimensions:

"DataFormat" for byte arrays includes class and length:

Datasets  (2)

Import names of all datasets in a file:

"Datasets" is the default HDF5 element:

Dimensions  (4)

Get dimensions of all datasets in a file:

Import dimensions of all datasets under a given group:

Dimensions of data with only a single element are indicated by an empty list:

Get dimensions of a specific dataset:

Specify the dataset using its index:

Groups  (1)

Import names of all groups in the file, with each group listed only once:

StructureGraph  (3)

The structure of a structured HDF5 file:

Structure of a flat HDF5 file:

Get a legend:

Summary  (1)

Get the file summary:

Export Elements  (28)

Attributes  (4)

Attach attributes to the root group:

Specify data format for attributes:

Export dataset's attributes both in "Datasets" and "Attributes" elements:

Each attribute must have a name that is unique among all attributes attached to the same object:

Try to export a different attribute with an already existing name:

It is possible to overwrite existing attributes with "AppendMode" set to "Overwrite":

Datasets  (13)

"Datasets" is the default export element. Data format and dimensions are automatically inferred from the expression:

Export expressions into different datasets, each of which can have a full path:

Inspect the structure of the HDF5 file:

Export a dataset with a custom data format:

Export a dataset with an attribute:

Create a scalar dataset with a single integer:

Create an array of integer numbers:

Export an array of real numbers:

Export a numeric array with complex numbers:

Note that complex numbers are exported to a compound dataset:

Use the "ComplexKeys" option to get complex numbers back:

Create a dataset with strings:

A ByteArray is stored as a "ByteArray" type (also called the opaque type in HDF5):

Associations with string keys are exported as elements of a compound type:

Create a dataset of initial size 2×2 that can later be extended to 10 rows and arbitrarily many columns:

Overwrite the first row with a list of 5 integers. This will extend the dataset to 5 columns:

Use the "SpaceSelection" subelement to overwrite the third and fourth columns:

Create a 3×4 dataset that can be arbitrarily extended:

Append a 3×3 array. This can only be done along the first dimension and Export will automatically detect that:

Sometimes data dimensions do not allow for automatic extensions of the dataset:

In this case, "SpaceSelection" can be specified manually:

Groups  (9)

HDF5 files always have a root group (with path "/"):

Export multiple groups by giving a list of paths:

Groups can have links to other groups or datasets:

Soft link names cannot contain the special character "/":

Soft link target must be a valid path, without "." or "..":

If a soft link target is missing, the link becomes a dangling link:

Create a file with two groups, a dataset and a hard link from one group to the dataset:

A hard link pointing to a nonexistent location cannot be added:

Create a new dataset under "/A/newDset":

Try to redirect the "newDset" link to point to group "B":

It is only possible with "Overwrite" mode:

Now the access to the second dataset is irreversibly lost. This is a resource leak as the dataset still occupies space in the file.

As with soft links, hard link name is always relative to the group path and must consist of exactly one path element:

The hard link target must be a valid path, without "." or "..":

NamedDatatypes  (2)

Export a simple type to the file:

Append a compound named data type to the same file:

Use the previously exported type when exporting a dataset:

Import data format and dimensions of the dataset:

Inspect exported data:

Named datatypes can carry attributes:

Import Options  (4)

"ComplexKeys"  (1)

HDF5 has no built-in type for complex numbers. Usually they are stored in a compound type:

Specify keys to be interpreted as real and imaginary parts of a complex number:

"TakeElements"  (3)

Import three points from a dataset:

This would be equivalent to but more efficient than extracting parts after importing the whole dataset:

Import a range of data from a dataset using spans:

Use suboptions for specifying offset, stride, and block:

Export Options  (8)

"AppendMode"  (1)

When appending data to an existing file using OverwriteTarget"Append", the export behavior for appending data can be specified using the "AppendMode" option. By default, "AppendMode""Extend" is used.

Create a simple file:

Add a new group:

Append a new dataset:

Append some attributes to "gr1/ds1" and a hard link from "gr1" to "gr2":

"AppendMode""Extend" does not allow for modifying existing attributes or links:

With "AppendMode""Overwrite", it is possible to overwrite existing objects and add new ones:

With "AppendMode""Overwrite", it is possible to modify data in existing datasets as long as data format and dimensions match:

To append new objects to the file with the guarantee that no existing structures in the file will be modified, use "AppendMode""Preserve":

ByteOrdering  (1)

HDF5 allows you to choose the byte ordering. Create a big-endian file using ByteOrdering-1:

Create a little-endian file:

The two files are actually different:

Import will work correctly independently of the byte ordering:

"ComplexKeys"  (3)

By default, complex numbers are exported to a compound type using "Re" and "Im" keys:

Specify different keys:

To import the data as complex numbers, specify the keys in Import:

Keys must be different:

OverwriteTarget  (3)

By default, with OverwriteTargetTrue, every call to Export writes a new file:

With OverwriteTargetFalse, the call to Export will fail if the output file already exists:

Export to certain formats, e.g. HDF5, supports the setting OverwriteTarget"Append", which makes Export act on the output file if it already exists, instead of overwriting it:

Properties & Relations  (1)

An association is, by default, exported as a compound dataset:

A list of rules is exported as multiple datasets:

Possible Issues  (1)

"Groups" lists each group in the file only once:

Notice that "/group2" is an unlisted valid path in the file:

Use an unlisted group path: