4. Writing Photon-HDF5 files

To create Photon-HDF5 files, users can convert existing files or save directly from suitably modified acquisition software. The conversion option is generally the simplest approach and, when using closed-source acquisition software, also the only one available (until vendors start supporting Photon-HDF5).

To simplify saving (and converting) Photon-HDF5 files we developed and maintain, phconvert, an open-source python library serving as reference implementation for the Photon-HDF5 format. While Photon-HDF5 can be created without phconvert, using only a HDF5 library, we recommend taking advantage of phconvert to simplify the writing step and to make sure that the saved file conforms to the specifications. Phconvert, in fact, checks that all mandatory fields are present and have correct names and types, and adds a description to each field. Phconvert can be directly used in programs written in Python or other languages that allow calling Python code (see next sections). phconvert permissive license (MIT) allows integration with both open and closed source software.

4.1. Converting files to Photon-HDF5

phconvert includes a browser-based interface using Jupyter Notebooks to convert vendor-specific file formats into Photon-HDF5 without requiring any python knowledge. The formats currently supported are HT3 (from PicoQuant TCSPC hardware), SPC/SET (from Becker & Hickl TCSPC hardware) as well as SM (a legacy file format developed by the WeissLab, UCLA).

We provide a demo service to run these notebooks online and convert one of these formats to Photon-HDF5 without software installation on the user’s computer.

Beyond the currently supported ones, other formats can be converted by writing a Python function to load the data and by using phconvert to save the data to Photon-HDF5. Taking the existing phconvert loader functions as examples, this task is relatively easy even for inexperienced Python programmers. See also the notebook Writing Photon-HDF5 files (view online).

We encourage interested users to contribute to load functions to phconvert so that out-of-the-box support for conversion of the largest number of formats can be provided. If you have an input file format not supported by phconvert please open a new issue on GitHub.

4.2. Save Photon-HDF5 from a third party-software

To directly save Photon-HDF5 files from within an acquisition software, there are several options. For programs written in Python, the obvious option is using phconvert which makes simple creating Photon-HDF5 files while assuring the validity of the output file. See for example the notebook Writing Photon-HDF5 files (view online).

For acquisition software written in other languages(e.g. C, MATLAB or LabVIEW), it is in principle possible to call python using the Python C API (see Embedding Python in Another Application). However understanding the Python C API requires a fairly good proficiency in C (and probably python).

In order to make it easy to create valid Photon-HDF5 in any language (without duplicating the effort of creating a library like phconvert in every language) we devised an alternative approach. The user can save the photon-data arrays (timestamps, detectors, nanotimes, etc…) in a plain HDF5 file. The remaining metadata is written in a simple text file (YAML). Next, a script called phforge reads the metadata and the photon-data arrays and creates a valid Photon-HDF5 file using phconvert. In this way, at the cost of a small inefficiency (writing some temporary files), a user can easily and reliably generate Photon-HDF5 files from any language. The metadata file is a text-based representation of the full Photon-HDF5 structure, excluding the photon-data arrays and some other field automatically filled by phconvert. To store this metadata we use YAML markup (a superset of JSON) for its simplicity and ability to describe hierarchical structures. For example, a minimal metadata file describing only mandatory fields is the following:

description: This is a dummy dataset which mimics smFRET data.

    num_pixels: 2                # using 2 detectors
    num_spots: 1                 # a single confocal excitation
    num_spectral_ch: 2           # donor and acceptor detection
    num_polarization_ch: 1       # no polarization selection
    num_split_ch: 1              # no beam splitter
    modulated_excitation: False  # CW excitation, no modulation
    lifetime: False              # no TCSPC in detection

        timestamps_unit: 10e-9   # 10 ns

To save the photon-data arrays the user needs to call the HDF5 library for the language of choice. For example, in MATLAB timestamps and detectors arrays can be saved with the following commands:

h5create('photon_data.h5', '/timestamps', size(timestamps), 'Datatype', 'int64')
h5write('photon_data.h5', '/timestamps', timestamps)
h5create('photon_data.h5', '/detectors', size(detectors), 'Datatype', 'uint8')
h5write('photon_data.h5', '/detectors', detectors)

Finally, once metadata and photon-data files have been saved, a Photon-HDF5 file can be created calling the phforge script as follows:

phforge metadata.yaml photon-data-arrays.h5 photon-hdf5-output.hdf5

Note that the file generated with this minimal metadata, does not contain the measurement_specs group which is in general necessary for a user to analyze the data.

The phforge script is available at http://photon-hdf5.github.io/phforge/. Examples of complete metadata files for all the supported measurement types are available at https://github.com/Photon-HDF5/phforge/tree/master/example_data.

A complete example of creating Photon-HDF5 files in LabVIEW using phforge can be found at https://github.com/Photon-HDF5/photon-hdf5-labview-write (for a MATLAB see next section).

Please use the mailing list if you have any questions.

4.3. Saving Photon-HDF5 from MATLAB

Creating Photon-HDF5 in MATLAB is easy using the approach described in the previous section, i.e. calling the script phforge.

Complete MATLAB examples can be found at https://github.com/Photon-HDF5/photon-hdf5-matlab-write.

In principle, it should be possible using a recent release of MATLAB (R2014b or later) to directly call python functions. Therefore it should be possible to directly call phconvert. However, in our recent attempt, we weren’t able to configure MATLAB in order to load the correct dynamic libraries (i.e. the HDF5 C library) required by phconvert.

4.4. Saving Photon-HDF5 from scratch using only an HDF5 library

To create Photon-HDF5 files from languages different than python the easiest option, by far, is calling the phforge script as described in previous section Save Photon-HDF5 from a third party-software.

If for some reason you cannot use phforge or phconvert, you have to implement routines to write Photon-HDF5 files using the HDF5 library for your platform, taking care of following the Photon-HDF5 specification. In the following paragraph we provide a few suggestions on how to proceed in this case.

To facilitate writing valid Photon-HDF5, we provide a JSON file containing all the official field names, a short description and a generic type definition (array, scalar, string or group). This JSON file can be used both to validate names and types of the data fields and to retrieve the standard short description (this is, in fact, what phconvert does). The developer needs to verify that all the mandatory fields are present. The description string should be saved for all the official fields in an attribute named “TITLE”. For compatibility with h5labview, we recommend to use a single-space string (” ”) for all the user fields that lack a description (phconvert uses this workaround too).

Furthermore, the /identity group should include the fields software_name and software_version to specify the name and the version of the software that created the file.

Finally, you can verify that generate files are compliant with the Photon-HDF5 specifications by using the phconvert function phconvert.hdf5.assert_valid_photon_hdf5_tables(). This function will raise errors or warnings if the input file does not follows the specs.