Easy navigation and data storage for HDF5

The hierarchical data format (HDF) is aimed to ensure efficient and equitable access to science and engineering data across platforms and environments. The h5py package provides a pythonic interface to the HDF5 binary data format and the h5io package simplifies this interface by introducing the read_hdf5() and write_hdf5() functions for loading and storing python objects in HDF5. The h5io package also provides a list_file_contents() function to print the internal, structure of an HDF5 file.

The h5io_browser package extends this interface by providing a pointer h5io_browser.Pointer to a specific path inside the hierarchical structure of the HDF5 file. With this pointer, data can be read, stored, copied and deleted from the HDF5 file, while at the same time simplifying the navigation inside the hierarchy of the file. The h5io_browser package is developed with three constraints and goals:

• Simplify navigating HDF5 files created by the h5io package. This includes interactive navigation inside an interactive Python shell or a Jupyter Notebook environment.
• Integrate standard functionality to interact with the data stored in the HDF5 file like read, write, copy and delete using the interface defined by the h5py package and the h5io package.
• Finally, balance flexibility and performance. Just like the h5io package, the h5io_browser only opens the HDF5 file when accessing the data and does not maintain an open file handle while waiting for user input. At the same time the interface defined by the h5io package is extended to store multiple python objects at the same time for improved performance.

Installation

The h5io_browser package can be installed either via the Python Package Index:

pip install h5io_browser

Or alternatively, via the community channel on the conda package manager maintained by the conda-forge community:

conda install -c conda-forge h5io_browser

Example

Demonstration of the basic functionality of the h5io_browser module.

Import Module

Start by importing the h5io_browser module:

import h5io_browser as hb

From the h5io_browser module the Pointer() object is created to access a new HDF5 file named new.h5:

hp = hb.Pointer(file_name="new.h5")

Write Data

For demonstration three different objects are written to the HDF5 file:

• a list with the numbers one and two is stored in the HDF5 path data/a_list
• an integer number is stored in the HDF5 path data/an_integer_number
• a dictionary is stored in the HDF5 path data/sub_path/a_dictionary

This can either be done using the edge notation, known from accessing python dictionaries, or alternatively using the write_dict() function which can store multiple objects in the HDF5 file, while opening it only once.

hp["data/a_list"] = [1, 2]
hp.write_dict(
    data_dict={
        "data/an_integer_number": 3,
        "data/sub_path/a_dictionary": {"d": 4, "e": 5},
    }
)

Read Data

One strength of the h5io_browser package is the support for interactive python environments like, Jupyter notebooks. To browse the HDF5 file by executing the Pointer() object:

hp

In comparison the string representation lists the file_name, h5_path as well as the nodes and groups at this h5_path:

str(hp)
>>> 'Pointer(file_name="/Users/jan/test/new.h5", h5_path="/") {"groups": ["data"], "nodes": []}'

List content of the HDF5 file at the current h5_path using the list_all() function:

hp.list_all()
>>> ['data']

In analogy the groups and nodes of any h5_path either relative to the current h5_path or as absolute h5_path can be analysed using the list_h5_path():

hp.list_h5_path(h5_path="data")
>>> {'groups': ['sub_path'], 'nodes': ['a_list', 'an_integer_number']}

To continue browsing the HDF5 file the edge bracket notation can be used, just like it s commonly used for python dictionaries to browse the HDF5 file:

hp["data"].list_all()
>>> ['a_list', 'an_integer_number', 'sub_path']

The object which is returned is again a Pointer with the updated h5_path, which changed from / to /data:

hp.h5_path, hp["data"].h5_path
>>> ('/', '/data')

Finally, individual nodes of the HDF5 file can be loaded with the same syntax using the / notation known from the file system, or by combining multiple edge brackets:

hp["data/a_list"], hp["data"]["a_list"]
>>> ([1, 2], [1, 2])

Convert to Dictionary

To computationally browse through the contents of an HDF5 file, the to_dict() method extends the interactive browsing capabilities. By default it returns a flat dictionary with the keys representing the h5_path of the individual nodes and the values being the data stored in these nodes. Internally, this loads the whole tree structure, starting from the current h5_path, so depending on the size of the HDF5 file this can take quite some time:

hp.to_dict()
>>> {'data/a_list': [1, 2],
>>>  'data/an_integer_number': 3,
>>>  'data/sub_path/a_dictionary': {'d': 4, 'e': 5}}

An alternative representation, is the hierarchical representation which can be enabled by the hierarchical being set to True. Then the data is represented as a nested dictionary:

hp.to_dict(hierarchical=True)
>>> {'data': {'a_list': [1, 2],
>>>   'an_integer_number': 3,
>>>   'sub_path': {'a_dictionary': {'d': 4, 'e': 5}}}}

With Statement

For compatibility with other file access methods, the h5io_browser package also supports the with statement notation. Still technically this does not change the behavior, even when opened with a with statement the HDF5 file is closed between individual function calls.

with hb.Pointer(file_name="new.h5") as hp:
    print(hp["data/a_list"])
>>> [1, 2]

Delete Data

To delete data from an HDF5 file using the h5io_browser the standard python del function can be used in analogy to deleting items from a python dictionary. To demonstrate the deletion a new node is added named data/new/entry/test:

hp["data/new/entry/test"] = 4

To list the node, the to_dict() function is used with the hierarchical parameter to highlight the nested structure:

hp["data/new"].to_dict(hierarchical=True)
>>> {'entry': {'test': 4}}

The node is then deleted using the del function. While this removes the node from the index the file size remains the same, which is one of the limitations of the HDF5 format. Consequently, it is not recommended to create and remove nodes in the HDF5 files frequently:

print(hp.file_size())
del hp["data/new/entry/test"]
print(hp.file_size())
>>> (18484, 18484)

Even after the deletion of the last node the groups are still included in the HDF5 file. They are not listed by the to_dict() function, as it recursively iterates over all nodes below the current h5_path:

hp["data/new"].to_dict(hierarchical=True)
>>> {}

Still with the list_all() function lists all nodes and groups at a current h5_path including empty groups, like the entry group in this case:

hp["data/new"].list_all()
>>> ['entry']

To remove the group from the HDF5 file the same del command is used:

del hp["data/new"]

After deleting both the newly created groups and their nodes the original hierarchy of the HDF5 file is restored:

hp.to_dict(hierarchical=True)
>>> {'data': {'a_list': [1, 2],
>>>  'an_integer_number': 3,
>>>  'sub_path': {'a_dictionary': {'d': 4, 'e': 5}}}}

Still even after deleting the nodes from the HDF5 file, the file size remains the same:

hp.file_size()
>>> 18484

Loop over Nodes

To simplify iterating recursively over all nodes contained in the selected h5_path the Pointer() object can be used as iterator:

hp_data = hp["data"]
{h5_path: hp_data[h5_path] for h5_path in hp_data}
>>> {'a_list': [1, 2],
>>>  'an_integer_number': 3,
>>>  'sub_path/a_dictionary': {'d': 4, 'e': 5}}

Copy Data

In addition to adding, browsing and removing data from an existing HDF5 file, the Pointer() object can also be used to copy data inside a given HDF5 file or copy data from one HDF5 file to another. A new HDF5 file is created, named copy.h5:

hp_copy = hb.Pointer(file_name="copy.h5")

The data is transferred from the existing Pointer() object to the new HDF5 file using the copy_to() functions:

hp["data"].copy_to(hp_copy)
hp_copy

Disclaimer

While we try to develop a stable and reliable software library, the development remains a opensource project under the BSD 3-Clause License without any warranties:

BSD 3-Clause License

Copyright (c) 2023, Jan Janssen
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

* Redistributions of source code must retain the above copyright notice, this
  list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above copyright notice,
  this list of conditions and the following disclaimer in the documentation
  and/or other materials provided with the distribution.

* Neither the name of the copyright holder nor the names of its
  contributors may be used to endorse or promote products derived from
  this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.