Saving and loading data in the HDF5 file format


HDF5 interface for the Julia language

Build Status Build status Coverage Status codecov

HDF5 is a file format and library for storing and accessing data, commonly used for scientific data. HDF5 files can be created and read by numerous programming languages. This package provides an interface to the HDF5 library for the Julia language.

Julia data (*.jld) and Matlab (*.mat) files

The core HDF5 functionality is the foundation for two special-purpose packages, used to read and write HDF5 files with specific formatting conventions. The first is the JLD ("Julia data") package, which implements a generic mechanism for reading and writing Julia variables. While one can use "plain" HDF5 for this purpose, the advantage of the JLD package is that it preserves the exact type information of each variable.

The other functionality provided through HDF5 is the ability to read and write Matlab *.mat files saved as "-v7.3". The Matlab-specific portions have been moved to Simon Kornblith's MAT.jl package.


pkg> add HDF5

If you run into installation errors, try to install the HDF5 library manually (version 1.8+ is required):

  • Debian/(K)Ubuntu: apt-get -u install hdf5-tools
  • OSX: brew install hdf5 (using Homebrew, please verify homebrew version is 1.8+ using brew --version)
  • Windows: Not recommended, as other HDF5 binaries may be compiled against a different C runtime from the Julia binary, which can cause Julia to crash when freeing memory allocated by libhdf5.

If you've installed the library but discover that Julia is not finding it, you can add the path to Julia's Libdl.DL_LOAD_PATH variable, e.g.,

push!(Libdl.DL_LOAD_PATH, "/opt/local/lib")

Inserting this command into your .juliarc.jl file will cause this to happen automatically each time you start Julia.

If you're on Linux but you do not have root privileges on your machine (and you can't persuade the sysadmin to install the libraries for you), you can download the binaries and place them somewhere in your home directory. To use HDF5, you'll have to start julia as

LD_LIBRARY_PATH=/path/to/hdf5/libs julia

You can set up an alias so this happens for you automatically each time you start julia.


Begin your code with

using HDF5

To read and write a variable to a file, one approach is to use the filename:

A = collect(reshape(1:120, 15, 8))
h5write("/tmp/test2.h5", "mygroup2/A", A)
data = h5read("/tmp/test2.h5", "mygroup2/A", (2:3:15, 3:5))

where the last line reads back just A[2:3:15, 3:5] from the dataset.

More fine-grained control can be obtained using functional syntax:

h5open("mydata.h5", "w") do file
    write(file, "A", A)  # alternatively, say "@write file A"

c = h5open("mydata.h5", "r") do file
    read(file, "A")

This allows you to add variables as they are generated to an open HDF5 file. You don't have to use the do syntax (file = h5open("mydata.h5", "w") works just fine), but an advantage is that it will automatically close the file (close(file)) for you, even in cases of error.

Julia's high-level wrapper, providing a dictionary-like interface, may also be of interest:

using HDF5

h5open("test.h5", "w") do file
    g = g_create(file, "mygroup") # create a group
    g["dset1"] = 3.2              # create a scalar dataset inside the group
    attrs(g)["Description"] = "This group contains only a single dataset" # an attribute

Convenience functions for attributes attached to datasets are also provided:

A = Vector{Int}(1:10)
h5write("bar.h5", "foo", A)
h5writeattr("bar.h5", "foo", Dict("c"=>"value for metadata parameter c","d"=>"metadata d"))
h5readattr("bar.h5", "foo")

Specific file formats

There is no conflict in having multiple modules (HDF5, JLD, and MAT) available simultaneously; the formatting of the file is determined by the open command.

Complete documentation

The HDF5 API is much more extensive than suggested by this brief introduction. More complete documentation is found in the doc directory.

The test directory contains a number of test scripts that also demonstrate usage.


  • Konrad Hinsen initiated Julia's support for HDF5

  • Tim Holy and Simon Kornblith (co-maintainers and primary authors)

  • Tom Short contributed code and ideas to the dictionary-like interface

  • Blake Johnson made several improvements, such as support for iterating over attributes

  • Isaiah Norton and Elliot Saba improved installation on Windows and OSX

  • Steve Johnson contributed the do syntax and Blosc compression

  • Mike Nolta and Jameson Nash contributed code or suggestions for improving the handling of HDF5's constants

  • Thanks also to the users who have reported bugs and tested fixes

First Commit


Last Touched

1 day ago


743 commits