Parallel HDF5 Configuration and Storage#
We extensively utilize Parallel-HDF5 data structure for configuration and simulation. We mainly use NeuroH5 package. Most of the descriptions/figures in the followingdiscussions are excerpted from [1]. For the full details of the tool, please refer the paper.
NeuroH5 Structure#
NeuroH5 implements data structure specialized for scalable and high-performance neuronal simulation. Based on parallel HDF5 format, the structure incorporates morphological, synaptic, and connectivity information from large neuronal network.
The NeuroH5 format focuses on two principal data structures:
graph projections: specify connections between two populations of cells
cell attributess: specify numerical attributes associated with individual cells
Hierarchical structure:
neuronal populations: a particular kind of neural species
attribute namespaces: logical grouping of attributes
global cell identifiers: individual cells within a population
cell attribute: homogeneous numerical vector that is contained iwthin a particular namespaces and associtated with a particular gid in a population
graph projection: collection of numerical vectors that specify the osurce and destination node indices of each edge, where each node index corresponds to a gid
Individual Cell Attribute#
The NeuroH5 data format for per-cell morphological, synaptic, connectivity, and other information of large neuronal network models, designed for further extension, construction, simulation, and analysis.
Each cell attributes contains:
cell index:
attribute pointer:
attribute value:
Dendritic Connectivity (Graph)#
Destination Block Sparse
Neuro IO#
Sample Structure#
