search

jonescompneurolab / hnn-core

Simulation and optimization of neural circuits for MEG/EEG source estimates
https://jonescompneurolab.github.io/hnn-core/
BSD 3-Clause "New" or "Revised" License
55 stars 52 forks source link

Support NetPyNE model specification #69

Open blakecaldwell opened 5 years ago

blakecaldwell commented 5 years ago

The code Salvador has written for HNN can replace the network-building code in network.py and cell.py (maybe feed.py). This would allow users to modify the network configuration in an ad hoc fashion. It also would integrate with the web-based HNN GUI (bokeh-based) being developed by MetaCell.

So the question is how can these 4 files be integrated into mne-neuron, and lead towards a single code base for HNN? https://github.com/jonescompneurolab/hnn/tree/netpyne/netpyne

jasmainak commented 5 years ago

I would start with a WIP pull request that makes netpyne an optional dependency (nested import) and shows how you can obtain the same result (with a test).

However, even before spending too much time doing this, I would suggest writing an example. How do you expect the user to interact with these new capabilities? If they change the network, in what ways would they do it in the context of EEG/MEG signals?

blakecaldwell commented 5 years ago

This is more work than I plan on undertaking. Instead, I may first adopt some of the changes in PR hnnsolver/mne-neuron#55 into the NetPyNE branch above.

If they change the network, in what ways would they do it in the context of EEG/MEG signals? 1) performance: it would allow for reconfiguration of the network (e.g. synaptic weights) without tearing it down (necessary to avoid memory leaks) and rebuilding it 2) it helps the HNN project goals by inferring a mechanistic understanding of EEG/MEG signals, not limited by the static network configuration used in HNN (and mne-python)

blakecaldwell commented 5 years ago

Oops, I meant mne-neuron instead of mne-python ^^

jasmainak commented 5 years ago

okay netpyne adoption is in my mental roadmap :) However, I don't want to rush it. Since the HNN code base is small it allows me to iterate fast, try things and learn more about Neuron. Adding a minimal GUI would not be that difficult (hopefully) once the code is a bit settled. In the meanwhile, I am hoping netpyne codebase improves a bit in terms of documentation / API etc. so one doesn't have to spend hours figuring out details of how it works.

blakecaldwell commented 4 years ago

Example of how this could look: https://github.com/jonescompneurolab/hnn-core/pull/93

This hides Neuron stuff inside of netpyne. Everything in hnn-core can be pickled, which makes it easier for joblibs and MPI to coexist.

Now network.py and cell.py just define the netpyne specification (not object-oriented) which is passed into netpyne. Work remains to make this more readable, and there are some questions about hard-coded values.

Note, several organizational splits: 1) Moved spiking functions out of Net and into their own class Spike 2) Moved simulation functions into simulation.py. These basically call netpyne simulation functions. 3) MPI features/complexity in mpi.py. This is for advanced use cases on cluster compute systems.

jasmainak commented 4 years ago

maybe we need a netpyne coding sprint to understand and improve netpyne ... :)

blakecaldwell commented 4 years ago

True. I think that'd be very helpful...