HDF5 support

The XML serializations of large NeuroML models can be prohibitive to store. For such cases, NeuroML also includes support for saving models in the binary HDF5 format via the NeuroMLHdf5Writer in libNeuroML. The same format can be exported also from the Java API (example).

The format of the export is documented below:

• Network is exported as a network HDF5 group with id, notes, and the temperature (optional) stored as attributes.

• Population is exported as a group with id population_<id of the population> with id, component, size, type, and property tags stored as attributes.

  • If the population is a population list that includes instances of cells, the locations of cells (x, y, z), these are stored in a 3 column table (“chunked array”) with a row per instance.

• Projection is exported as a group with id project_<id of the projection> with id, type, presynapticPopulation, postSynapticPopulation, synapse as attributes.

  • Connection and ConnectionWD elements in projections are stored as rows in a table with the first two columns as the pre_cell_id and post_cell_id respectively, and the successive columns for the necessary attributes.

• ElectricalProjection is exported similar to Projection with the ElectricalConnection, ElectricalConnectionInstance, and ElectricalConnectionInstanceW entries stored in tables.

• ContinuousProjection is exported similar to Projection with the ContinuousConnection, ContinuousConnectionInstance, and ContinuousConnectionInstanceW entries stored in tables.

• InputList is exported similar to Projection with the Input, and InputW entries stored in tables.

For more details, the source code of these export functions can be seen here in the libNeuroML repository and here in org.neuroml.model.

HDF5 NeuroML files can be read and processed by jnml and pynml in the same way as XML files (see here for LEMS Simulation file examples which reference HDF5 NeuroML models).