Jupyter Notebook for NeuroML model using `info` method

[irahulsonkar]

Generic notebook for loading existing NeuroML models and run interactively.

Current Status:

• loads any NeuroML model using libneuroml's info method and read_neuroml2_file()
• all details read from the model are arranged in a structured way using jupyter widgets (mainly tabs and accordions)
• the model can be written to file (NeuroMLProject.nml) using Update Model button in notebook which calls writers.NeuroMLWriter.write()
• the Validate Model button can be used to validate the file against NeuroML2 schema using pynml.validate_neuroml2()
• model can be run using defaults settings/parameters available in the NeuroML model by clicking Run NeuroML button in notebook which calls pynml.run_lems_with_jneuroml()
• outputs can be plotted interactively with matplotlib widget using Plot Outptut button

Test Models: 1. Hodgkin Huxley Tutorial: LEMS - https://github.com/openworm/hodgkin_huxley_tutorial/blob/master/Tutorial/Source/LEMS_HH_Simulation.xml Network - https://github.com/openworm/hodgkin_huxley_tutorial/blob/master/Tutorial/Source/HHCellNetwork.net.nml

2. BahlEtAl2012_ReducedL5PyrCell: LEMS - https://github.com/OpenSourceBrain/BahlEtAl2012_ReducedL5PyrCell/blob/master/NeuroML2/LEMS_pyr_single_comp.xml Network - https://github.com/OpenSourceBrain/BahlEtAl2012_ReducedL5PyrCell/blob/master/NeuroML2/pyr_soma.net.nml

3. AllenInstituteNeuroML: LEMS - https://github.com/OpenSourceBrain/AllenInstituteNeuroML/blob/master/CellTypesDatabase/models/NeuroML2/LEMS_SomaTest.xml Network - https://github.com/OpenSourceBrain/AllenInstituteNeuroML/blob/master/CellTypesDatabase/models/NeuroML2/Soma_Test.net.nml

Future Work:

• LEMS data loaded in notebook under LEMS tab needs to be organised for better visuals (child objects can be made subset based on parent_id)
• editing model parameters using widget boxes, presently widgets are editable for values but its not written to the neuroml file when clicking Update Model button since that writes the values from loaded neuroml object in python and not from the widget values
• easy addition of new parameters to the loaded NeuroML model using widgets (e.g. adding new pluse generation by clicking some buttons and entering the fields)

[sanjayankur31]

Thanks @irahulsonkar . Looks good. Tested it out and works well too. :+1:

[sanjayankur31]

• I think another future work is to generalise all of this in the same way as nwbwidgets. The idea is that there will be a nml2widget(nml_model) function that will take any NeuroML class object (python object) and generate widgets, sort of as a general front end to info(). So users can use it to inspect individual components, or the whole document (for larger models, the whole document may not work---I tried it with one and it was taking forever so I had to kill it). We can then leave the reading of the file etc. to the user---as long as it's a NeuroML object, it will be converted to a/multiple widgets. This way, it doesn't have to be a NeuroML file that is read in---it can be me writing my model in NeuroML using Python, and then calling the nml2widget function later to see it all (similar to how we use info).

• While we modularise this, it'll be good to split out nm2widget and lemssimulation2widget because users may only want to inspect the model, and not the associated lems simulation or vice-versa.

• similar to nwbwidgets, we can then create "dashboards" (and allow users to create dashboards for their projects). The default can be what you have here---shows everything in a nice structured way

• as you've noted, the LEMS bits need a little tweaking.

• possible improvement: replace the text boxes with file selectors if possible (just so that users can use a portal to select the file instead of having to write the path there manually)

So, this is all a very very useful starting point. I think we'll create a new python package, nmlwidgets or something, and refactor the code here to follow the nwbwidgets model.

[irahulsonkar]

@sanjayankur31 Thanks for the review and nice suggestions.

• Regarding the first point, taking any NeuroML class object and generating widget. It is already there in the notebook as an example, I think in the last cell (Exploring Sub-Models) where you can pass any python object on which info works, say pulse generator or ion channel.

• The notebook by default passes full NeuroML class object which is read from the file but the user can modify it similar to example in last cells, presently commented out.

  runner.createAccordions(nml_doc.networks[0],'Networks'))    #pass the object and a tittle for parent accordion

  This will create only accordions for the given object. Similarly, if user wants first level as tabs then accordions following can be used, (probably this function should be renamed something to createTabWithAccordions())

  runner.generateDashboard(nml_doc.networks[0])    #pass the object (tittle not required)

• Yes, loading full nml_doc for larger models gets very slow. I was thinking to include a progress bar so the user atleast know its running and loading models.

• For replacing the text boxes with file selectors, I planned to used FileUpload widget of jupyter-notebook but during testing/development phase i found text boxes to be easier/faster, maybe later you can consider using this kind of widget.

[sanjayankur31]

I'll merge this now. Thanks again @irahulsonkar

[sanjayankur31]

Edit: can't yet, dont have permissions, will merge when we have that sorted out :)