HDF5 interface for the Julia language

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 modules, used to read and write HDF5 files with specific formatting conventions. The first is the JLD ("Julia data") module (provided in this 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 module 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 repository.

Installation

Within Julia, use the package manager:

Pkg.add("HDF5")

You also need to have the HDF5 library installed on your system (version 1.8 or higher is required), but for most users no additional steps should be required; the HDF5 library should be installed for you automatically when you add the package.

If you have to install the HDF5 library manually, here are some examples of how to do it:

• Debian/(K)Ubuntu: apt-get -u install hdf5-tools
• OSX: brew tap homebrew/science; brew install hdf5 (using Homebrew)
• Windows: It is highly recommended that you use the HDF5 library fetched by this package. Other HDF5 binaries may be compiled against a different C runtime from the Julia binary, which will 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 Sys.DL_LOAD_PATH variable, e.g.,

push!(Sys.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.

Quickstart

To use the JLD module, begin your code with

using HDF5, JLD

If you just want to save a few variables and don't care to use the more advanced features of HDF5, then a simple syntax is:

t = 15
z = [1,3]
save("/tmp/myfile.jld", "t", t, "arr", z)

Here we're explicitly saving t and z as "t" and "arr" within myfile.jld. You can alternatively pass save a dictionary; the keys must be strings and are saved as the variable names of their values within the JLD file. You can read these variables back in with

d = load("/tmp/myfile.jld")

which reads the entire file into a returned dictionary d. Or you can be more specific and just request particular variables of interest. For example, z = load("/tmp/myfile.jld", "arr") will return the value of arr from the file and assign it back to z.

There are also convenience macros @save and @load that work on the variables themselves. @save "/tmp/myfile.jld" t z will create a file with just t and z; if you don't mention any variables, then it saves all the variables in the current module. Conversely, @load will pop the saved variables directly into the global workspace of the current module. However, keep in mind that these macros have significant limitations: for example, you can't use @load inside a function, there are constraints on using string interpolation inside filenames, etc. These limitations stem from the fact that Julia compiles functions to machine code before evaluation, so you cannot introduce new variables at runtime or evaluate expressions in other workspaces. The save and load functions do not have these limitations, and are therefore recommended as being considerably more robust, at the cost of some slight reduction of convenience.

For plain HDF5 files, you can similarly say

A = 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:

jldopen("mydata.jld", "w") do file
    write(file, "A", A)  # alternatively, say "@write file A"
end

c = jldopen("mydata.jld", "r") do file
    read(file, "A")
end

This allows you to add variables as they are generated to an open JLD file. You don't have to use the do syntax (file = jldopen("mydata.jld", "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. This is demonstrated with the "plain" (unformatted) HDF5 interface:

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
end

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

More extensive documentation is found in the doc/ directory.

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

Credits

• 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, and string->type conversion in the JLD module

• 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

• 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