neuroml / org.neuroml.model Goto Github PK

I was doing some profiling of Geppetto to see where the bulk of the time was spent with NeuroML documents, I found the following:

With the Purkinje cell:

[2015-02-07 15:02:54.637] INFO  http-bio-8080-exec-4         o.geppetto.model.neuroml.services.NeuroMLModelInterpreterService  Parsed NeuroML document of size 1170KB, took 1554ms

With the Auditory cortex model:

[2015-02-07 15:06:10.012] INFO  http-bio-8080-exec-5         o.geppetto.model.neuroml.services.NeuroMLModelInterpreterService  Parsed NeuroML document of size 2287KB, took 19064ms 

Auditory is twice the size and takes ten times the time to parse.
Twenty seconds is a LOT (even one second is).

This is the code

            /*
             * NEUROML
             */
            start=System.currentTimeMillis();
            NeuroMLConverter neuromlConverter = new NeuroMLConverter();
            NeuroMLDocument neuroml = neuromlConverter.loadNeuroML(neuromlString);
            _logger.info("Parsed NeuroML document of size "+neuromlString.length()/1024+"KB, took "+(System.currentTimeMillis()-start)+"ms");

Using version 1.4.0. but pretty sure it's the same with any version. This is with single selfcontained neuroml files that contain all the inclusion inside them.
cc @pgleeson @adrianq @borismarin

e.g. see
https://bitbucket.org/mbezaire/nc_superdeep/src/fab4d82bae57a42a7da0e595d0cf04f203787295/neuroConstruct/generatedNeuroML2/pvbasketcell.cell.nml?at=default

segmentGroups marked with neuroLexId sao864921383, e.g.

<segmentGroup id="bcdendBAS2_2" neuroLexId="sao864921383">
<member segment="39"/>
<member segment="40"/>
</segmentGroup>

can be used to specify a subset of segmentGroups in a cell which contain unbranched (each segment has a unique parent in this group apart from the first) and non overlapping (every segment in cell belongs to one and only one of these segmentGroups) groups.

A test for this should be added to https://github.com/NeuroML/org.neuroml.model/blob/development/src/main/java/org/neuroml/model/util/NeuroML2Validator.java

e.g. https://github.com/NeuroML/org.neuroml.model/actions/runs/3525678094/jobs/5913288286

When parsing a neuroml doc with jlems, presynapticPopulation and postsynapticPopulation in projection component should be refcomponents. For instance, synapse attribute in a projection component is parse as a refcomponent. However pre and post synaptic population are just attributes.

@pgleeson I have created this issue in this repo as I guess it has to do with neuroml comp definition but maybe it makes more sense to move it to another repo.

Some exceptions I am finding in the log. Doesn't happen all the time.

In https://github.com/NeuroML/org.neuroml.model/blob/development/src/main/java/org/neuroml/model/util/NeuroML2Validator.java

• Ensure existence of imported files
• Ensure name of comps & populations doesn't start with number - move to schema...
• Projection - check existence of pop id
• Check population/populationList & existence of instances
• Connection: check existence of seg in pre/post pop comp
• Connection: parse pre/postCellId string for valid pop id & pop index

nc_ca1 h5 file gives

Error reading request, ignored java.lang.ArrayIndexOutOfBoundsException: 3
        at org.neuroml.model.util.hdf5.NetworkHelper.getLocation(NetworkHelper.java:178)
        at org.geppetto.model.neuroml.modelInterpreterUtils.PopulateTypes.createPopulationTypeVariable(PopulateTypes.java:565)

this is fixed in org.neuroml.model development but awaiting release

numberInternalDivisions, nseg in NEURON, was recommended for inclusion as a property subelement of annotation in segmentGroup. Some APIs have trouble reading what's inside annotation (because it's defined as <xs:any.. > in the Schema), so now it's recommended to put it outside annotation:

<segmentGroup id = "mainDendSec"  neuroLexId="sao864921383">
    <property tag="numberInternalDivisions" value="3"/>
....

Validator should check for preferred use & possible inconsistency