Representing spatial structure and calculating distancesΒΆ

The space module contains classes for specifying the locations of neurons in space and for calculating the distances between them.

Neuron positions can be defined either manually, using the positions attribute of a Population or using a Structure instance which is passed to the Population constructor.

A number of different structures are available in space. It is simple to define your own Structure sub-class if you need something that is not already provided.

The simplest structure is a grid, whether 1D, 2D or 3D, e.g.:

 >>> from pyNN.space import *
 >>> line = Line(dx=100.0, x0=0.0, y=200.0, z=500.0)
 >>> line.generate_positions(7)
 array([[   0.,  100.,  200.,  300.,  400.,  500.,  600.],
        [ 200.,  200.,  200.,  200.,  200.,  200.,  200.],
        [ 500.,  500.,  500.,  500.,  500.,  500.,  500.]])
 >>> grid = Grid2D(aspect_ratio=3, dx=10.0, dy=25.0, z=-3.0)
 >>> grid.generate_positions(3)
 array([[  0.,  10.,  20.],
        [  0.,   0.,   0.],
        [ -3.,  -3.,  -3.]])
 >>> grid.generate_positions(12)
 array([[  0.,   0.,  10.,  10.,  20.,  20.,  30.,  30.,  40.,  40.,  50.,  50.],
        [  0.,  25.,   0.,  25.,   0.,  25.,   0.,  25.,   0.,  25.,   0.,  25.],
        [ -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.,  -3.]])

Here we have specified an x:y ratio of 3, so if we ask the grid to generate positions for 3 neurons, we get a 3x1 grid, 12 neurons a 6x2 grid, 27 neurons 9x3, etc.

BY default, grid positions are filled sequentially, iterating first over the z dimension, then y, then x, but we can also fill the grid randomly:

>>> rgrid = Grid2D(aspect_ratio=1, dx=10.0, dy=10.0, fill_order='random', rng=NumpyRNG(seed=13886))
>>> rgrid.generate_positions(9)
array([[ 10.,  10.,  10.,   0.,   0.,   0.,  20.,  20.,  20.],
       [  0.,  10.,  20.,  10.,  20.,   0.,   0.,  10.,  20.],
       [  0.,   0.,   0.,   0.,   0.,   0.,   0.,   0.,   0.]])

The space module also provides the RandomStructure class, which distributes neurons randomly and uniformly within a given volume:

 >>> glomerulus = RandomStructure(boundary=Sphere(radius=200.0), rng=NumpyRNG(seed=34534))
 >>> glomerulus.generate_positions(5)
 array([[ -19.78455022,   33.21412264,  -79.4314059 ,  143.39033263,  -63.18242977],
        [  56.17281502,  -23.15159309,  131.89071845,  -73.73583484,  -8.86422999],
        [ -78.88348228,   -3.97408513,  -95.03056844,   45.13969087,  -111.67070498]])

The volume classes currently available are Sphere and Cuboid.

Defining your own Structure classes is straightforward, just inherit from BaseStructure and implement a generate_positions() method:

class MyStructure(BaseStructure):
    parameter_names = ("spam", "eggs")

    def __init__(self, spam=3, eggs=1):
        ...

    def generate_positions(self, n):
        ...
        # must return a 3xn numpy array

To define your own Shape class for use with RandomStructure, subclass Shape and implement a sample() method:

class Tetrahedron(Shape):

    def __init__(self, side_length):
        ...

    def sample(self, n, rng):
       ...
       # return a nx3 numpy array.

Note

rotation of structures is currently missing, but is planned for a future release.