v3.3.0

This release updates the hdf imports to remove the dependency on HDFview.
On unix, you might need to install jhdf5 and javahdf5.

Either git clone or download stochdiff-master.zip, do NOT use the other two zip files from 2018.

To compile and build jar file, type:
util/hdf5-maven-stub.sh 1.12.1
mvn package

NOTE: 1.12.1 is the java-hdf5 version. Use a different argument to hdf5-maven-stub.sh if your version is different

v3.2.4

This release updates the hdf imports (remove ncsa) to work with HDFview-2.1.4 and jarhdf5-3.3.2. It was compiled with eclipse, and contains the required libraries. For HDF5 output on Windows, user must install HDFview2.1.4 - see README for more information.

On unix, you might need to install jhdf5. On fedora, type:
sudo dnf install jhdf5

v3.2.3

This release adds a special constraint to treat any injection event as leap barrier for the whole voxel. This is a work-around for the case where injections into a given voxel influence events in the same voxel which are not directly dependent, but only dependent through an intermediary event.

v3.2.2

Bionomial and Poisson number generators are completely rewritten. Our own custom code which used interpolated tables to generate bionomial numbers is removed, and new unbiased generators are used. This should be both faster and more accurate. Configurability of the number generators is disabled: we always use Poisson random numbers for the 0th order, and Binomial random numbers for higher orders.

This release fixes a bug introduced in v3.2.1: for any species which are not present in the ConcentrationSet for some region, we now again fall back to the default ConcentrationSet.

v3.2.1

For injections which target a group of spines (e.g. injectionSite="sa2[1:3].pointD"), the rate is now divided by the number of spines in the group. This means that the specified rate corresponds directly to the number of injected molecules, and does not change if the number of spines in the group changes. This makes injection into spines behave similarly to injections into the submembrane (e.g. `injectionSite="dendrite:submembrane"), which also does not change the number of injected molecules when the morphology is changed to have more or less volumes.

It is now possible to disable diffusion, reactions, or injections from the command line:

java neurord.StochDiff -v model.xml -Dneurord.diffusion=false

This might be useful for debugging and benchmarking.

If a SurfaceDensitySet is specified without a region attribute, it will be used as the default for regions which don't have their SurfaceDensitySet specified. This matches the behaviour for ConcentrationSet.

Build timestamp in --help output now includes the time zone.

A bug which caused some molecules to be initialized improperly (with 0 initial concentration for a single region) was fixed.

v3.2

A significant bug in propensity calculation is fixed: propensities were not multiplied by 0.602 (the conversion constant from concentrations to particle numbers). In effect, all reactions were too fast.

HDF5 format is slightly changed:

• various metadata are saved as attributes, not as separate tables
• compression is enabled for all arrays, this gives large space savings
• shuffle filter is enabled, giving more space savings

Python support code can now display the diff between configuration of two .h5 files, or an .h5 file and an .xml file, or two .xml files. XML includes are automatically processed.
Molecule names can be specified using globs.

v3.1.4

• time is reported in human readable units (weeks, days, hours, minutes, seconds…)
• a crash when writing statistics and no statistics were to be written is fixed

v3.1.3

• add descriptions for all statistics rows
• update the python module documentation for recent changes
• fix various synchronization bugs when running multiple trials
• allow units to be specified using unicode

v3.1.2

• New options to collect statistics (-s injections) and set simulation time (--runtime XXX).
• Slightly more efficient writing of HDF5 results.

v3.1.1

NeuroRD 3.1.1