Groups are the container mechanism by which HDF5 files are organized. From a Python perspective, they operate somewhat like dictionaries. In this case the “keys” are the names of group members, and the “values” are the members themselves (Group and Dataset) objects.

Group objects also contain most of the machinery which makes HDF5 useful. The File object does double duty as the HDF5 root group, and serves as your entry point into the file:

>>> f = h5py.File('foo.hdf5','w')
>>> list(f.keys())

Names of all objects in the file are all text strings (str). These will be encoded with the HDF5-approved UTF-8 encoding before being passed to the HDF5 C library. Objects may also be retrieved using byte strings, which will be passed on to HDF5 as-is.

Creating groups

New groups are easy to create:

>>> grp = f.create_group("bar")
>>> subgrp = grp.create_group("baz")

Multiple intermediate groups can also be created implicitly:

>>> grp2 = f.create_group("/some/long/path")
>>> grp3 = f['/some/long']


class h5py.Group(identifier)

Generally Group objects are created by opening objects in the file, or by the method Group.create_group(). Call the constructor with a GroupID instance to create a new Group bound to an existing low-level identifier.


Iterate over the names of objects directly attached to the group. Use Group.visit() or Group.visititems() for recursive access to group members.


Dict-like membership testing. name may be a relative or absolute path.


Retrieve an object. name may be a relative or absolute path, or an object or region reference. See Dict interface and links.

__setitem__(name, value)

Create a new link, or automatically create a dataset. See Dict interface and links.


Check that the group is accessible. A group could be inaccessible for several reasons. For instance, the group, or the file it belongs to, may have been closed elsewhere.

>>> f =
>>> group = f["MyGroup"]
>>> f.close()
>>> if group:
...     print("group is accessible")
... else:
...     print("group is inaccessible")
group is inaccessible

Get the names of directly attached group members. Use Group.visit() or Group.visititems() for recursive access to group members.


set-like object.


Get the objects contained in the group (Group and Dataset instances). Broken soft or external links show up as None.


a collection or bag-like object.


Get (name, value) pairs for object directly attached to this group. Values for broken soft or external links show up as None.


a set-like object.

get(name, default=None, getclass=False, getlink=False)

Retrieve an item, or information about an item. name and default work like the standard Python dict.get.

  • name – Name of the object to retrieve. May be a relative or absolute path.

  • default – If the object isn’t found, return this instead.

  • getclass – If True, return the class of object instead; Group or Dataset.

  • getlink – If true, return the type of link via a HardLink, SoftLink or ExternalLink instance. If getclass is also True, returns the corresponding Link class without instantiating it.


Recursively visit all objects in this group and subgroups. You supply a callable with the signature:

callable(name) -> None or return value

name will be the name of the object relative to the current group. Return None to continue visiting until all objects are exhausted. Returning anything else will immediately stop visiting and return that value from visit:

>>> def find_foo(name):
...     """ Find first object with 'foo' anywhere in the name """
...     if 'foo' in name:
...         return name
>>> group.visit(find_foo)

Recursively visit all objects in this group and subgroups. Like Group.visit(), except your callable should have the signature:

callable(name, object) -> None or return value

In this case object will be a Group or Dataset instance.

These methods are like visit() and visititems(), but work on the links in groups, rather than the objects those links point to. So if you have two links pointing to the same object, these will ‘see’ both. They also see soft & external links, which visit() and visititems() ignore.

The second argument to the callback for visititems_links is an instance of one of the link classes.

New in version 3.11.

move(source, dest)

Move an object or link in the file. If source is a hard link, this effectively renames the object. If a soft or external link, the link itself is moved.

  • source (String) – Name of object or link to move.

  • dest (String) – New location for object or link.

copy(source, dest, name=None, shallow=False, expand_soft=False, expand_external=False, expand_refs=False, without_attrs=False)

Copy an object or group. The source can be a path, Group, Dataset, or Datatype object. The destination can be either a path or a Group object. The source and destination need not be in the same file.

If the source is a Group object, by default all objects within that group will be copied recursively.

When the destination is a Group object, by default the target will be created in that group with its current name (basename of You can override that by setting “name” to a string.

  • source – What to copy. May be a path in the file or a Group/Dataset object.

  • dest – Where to copy it. May be a path or Group object.

  • name – If the destination is a Group object, use this for the name of the copied object (default is basename).

  • shallow – Only copy immediate members of a group.

  • expand_soft – Expand soft links into new objects.

  • expand_external – Expand external links into new objects.

  • expand_refs – Copy objects which are pointed to by references.

  • without_attrs – Copy object(s) without copying HDF5 attributes.

create_group(name, track_order=None)

Create and return a new group in the file.

  • name (String or None) – Name of group to create. May be an absolute or relative path. Provide None to create an anonymous group, to be linked into the file later.

  • track_order – Track dataset/group/attribute creation order under this group if True. Default is h5.get_config().track_order.


The new Group object.


Open a group in the file, creating it if it doesn’t exist. TypeError is raised if a conflicting object already exists. Parameters as in Group.create_group().

create_dataset(name, shape=None, dtype=None, data=None, **kwds)

Create a new dataset. Options are explained in Creating datasets.

  • name – Name of dataset to create. May be an absolute or relative path. Provide None to create an anonymous dataset, to be linked into the file later.

  • shape – Shape of new dataset (Tuple).

  • dtype – Data type for new dataset

  • data – Initialize dataset to this (NumPy array).

  • chunks – Chunk shape, or True to enable auto-chunking.

  • maxshape – Dataset will be resizable up to this shape (Tuple). Automatically enables chunking. Use None for the axes you want to be unlimited.

  • compression – Compression strategy. See Filter pipeline.

  • compression_opts – Parameters for compression filter.

  • scaleoffset – See Scale-Offset filter.

  • shuffle – Enable shuffle filter (T/F). See Shuffle filter.

  • fletcher32 – Enable Fletcher32 checksum (T/F). See Fletcher32 filter.

  • fillvalue – This value will be used when reading uninitialized parts of the dataset.

  • track_times – Enable dataset creation timestamps (T/F).

  • track_order – Track attribute creation order if True. Default is h5.get_config().track_order.

  • external – Store the dataset in one or more external, non-HDF5 files. This should be an iterable (such as a list) of tuples of (name, offset, size) to store data from offset to offset + size in the named file. Each name must be a str, bytes, or os.PathLike; each offset and size, an integer. The last file in the sequence may have size h5py.h5f.UNLIMITED to let it grow as needed. If only a name is given instead of an iterable of tuples, it is equivalent to [(name, 0, h5py.h5f.UNLIMITED)].

  • allow_unknown_filter – Do not check that the requested filter is available for use (T/F). This should only be set if you will write any data with write_direct_chunk, compressing the data before passing it to h5py.

  • rdcc_nbytes – Total size of the dataset’s chunk cache in bytes. The default size is 1024**2 (1 MiB).

  • rdcc_w0 – The chunk preemption policy for this dataset. This must be between 0 and 1 inclusive and indicates the weighting according to which chunks which have been fully read or written are penalized when determining which chunks to flush from cache. A value of 0 means fully read or written chunks are treated no differently than other chunks (the preemption is strictly LRU) while a value of 1 means fully read or written chunks are always preempted before other chunks. If your application only reads or writes data once, this can be safely set to 1. Otherwise, this should be set lower depending on how often you re-read or re-write the same data. The default value is 0.75.

  • rdcc_nslots – The number of chunk slots in the dataset’s chunk cache. Increasing this value reduces the number of cache collisions, but slightly increases the memory used. Due to the hashing strategy, this value should ideally be a prime number. As a rule of thumb, this value should be at least 10 times the number of chunks that can fit in rdcc_nbytes bytes. For maximum performance, this value should be set approximately 100 times that number of chunks. The default value is 521.

require_dataset(name, shape, dtype, exact=False, **kwds)

Open a dataset, creating it if it doesn’t exist.

If keyword “exact” is False (default), an existing dataset must have the same shape and a conversion-compatible dtype to be returned. If True, the shape and dtype must match exactly.

If keyword “maxshape” is given, the maxshape and dtype must match instead.

If any of the keywords “rdcc_nslots”, “rdcc_nbytes”, or “rdcc_w0” are given, they will be used to configure the dataset’s chunk cache.

Other dataset keywords (see create_dataset) may be provided, but are only used if a new dataset is to be created.

Raises TypeError if an incompatible object already exists, or if the shape, maxshape or dtype don’t match according to the above rules.


exact – Require shape and type to match exactly (T/F)

create_dataset_like(name, other, **kwds)

Create a dataset similar to other, much like numpy’s _like functions.

  • name – Name of the dataset (absolute or relative). Provide None to make an anonymous dataset.

  • other – The dataset whom the new dataset should mimic. All properties, such as shape, dtype, chunking, … will be taken from it, but no data or attributes are being copied.

Any dataset keywords (see create_dataset) may be provided, including shape and dtype, in which case the provided values take precedence over those from other.

create_virtual_dataset(name, layout, fillvalue=None)

Create a new virtual dataset in this group. See Virtual Datasets (VDS) for more details.

  • name (str) – Name of the dataset (absolute or relative).

  • layout (VirtualLayout) – Defines what source data fills which parts of the virtual dataset.

  • fillvalue – The value to use where there is no data.


Assemble a virtual dataset in this group.

This is used as a context manager:

with f.build_virtual_dataset('virt', (10, 1000), np.uint32) as layout:
    layout[0] = h5py.VirtualSource('foo.h5', 'data', (1000,))

Inside the context, you populate a VirtualLayout object. The file is only modified when you leave the context, and if there’s no error.

  • name (str) – Name of the dataset (absolute or relative)

  • shape (tuple) – Shape of the dataset

  • dtype – A numpy dtype for data read from the virtual dataset

  • maxshape (tuple) – Maximum dimensions if the dataset can grow (optional). Use None for unlimited dimensions.

  • fillvalue – The value used where no data is available.


Attributes for this group.


The groups’s low-level identifier; an instance of GroupID.


An HDF5 object reference pointing to this group. See Using object references.


A proxy object allowing you to interrogate region references. See Using region references.


String giving the full path to this group.


File instance in which this group resides.


Group instance containing this group.