3D localization of neurons in HNN

[ntolley]

I've been working on modifying how neurons in the model are positioned in 3D space and running into some points that could be improved or clarified:

• The default cell separation is currently way too small (1 micron), perhaps net.set_cell_positions(inplane_distance=10) is more appropriate?
• What is the rationale of having space_constant=3.0 in the evoked drive? Is this
• Related to the above, it seems like there should be an option to simply set a fixed weight with no spatial dependency. The best we can do right now is to set the lamtha/space constant to a really huge number
• Since the evoked drive weights vary with space, the location of the artificial cell matters quite a lot when you start changing the network.

Also we need to set some sort of tolerance/threshold on the distance dependent delay. The delay is calculated as follows:

delay = nc_dict['A_delay'] / _calculate_gaussian(cell_dist, 1, scaled_lamtha)

When the cell dist is large, the gaussian evaluates to approximately zero, which produces an overflow when we calculate its reciprocal. Maybe the solution is to set the delay to nan or something and just remove the offending connections?

[jasmainak]

I propose that net.cell_cell_positions should have an additional parameter scale_lamtha which is True by default. It will go over all the connections in the network and scale the lamtha by the same ratio as the inplane distance. Wdyt?

At the moment, I am not sure what the units of lamtha are? (we claim to work with physical units)

I would advocate against hackish solutions like using nan when the computation overflows

[rythorpe]

My understanding is that lamtha currently represents the connection weight between neurons based on their distance in indices (e.g., lamtha = 1.0 means that the synaptic weight falls off to 1/e at one grid position away) and that inplane_distance only applies to extracellular recordings.

[jasmainak]

We could have an option lamtha=None which makes the weights uniform ...

[rythorpe]

+1