How to replace call to sp._getNeighborsND()

rhyolight · November 4, 2016, 9:55pm

@Victor @mrcslws @ycui I used to have this logic running to show inhibition masks over topological input.

htm-community/nupic-history-server/blob/master/nupic_history/sp_facade.py#L411-L414


import uuid
import multiprocessing
import time


import numpy as np


from nupic_history import SpSnapshots as SNAPS
from nupic_history.utils import compressSdr
from nupic.math import topology




class SpFacade(object):


  def __init__(self, sp, ioClient, iteration=None):
    """
    A wrapper around the HTM Spatial Pooler that can save SP state to Redis for
    each compute cycle. Adds a "save=" kwarg to compute().


    :param sp: Either an instance of Spatial Pooler or a string model id
    :param ioClient: Instantiated IO client

This file has been truncated. show original

Can you tell me how to get this now that _getNeighborsND() is gone?

mrcslws · November 5, 2016, 2:47am

If it’s using wrapAround, use wrappingNeighborhood

github.com

numenta/nupic/blob/6924a904644bbb8cdba6ca7a62050cf8456614f8/src/nupic/math/topology.py#L122


intervals = []
for i, dimension in enumerate(dimensions):
  left = max(0, centerPosition[i] - radius)
  right = min(dimension - 1, centerPosition[i] + radius)
  intervals.append(xrange(left, right + 1))


coords = numpy.array(list(itertools.product(*intervals)))
return numpy.ravel_multi_index(coords.T, dimensions)




def wrappingNeighborhood(centerIndex, radius, dimensions):
"""
Like 'neighborhood', except that the neighborhood isn't truncated when it's
near an edge. It wraps around to the other side.


@param centerIndex (int)
The index of the point. The coordinates are expressed as a single index by
using the dimensions as a mixed radix definition. For example, in dimensions
42x10, the point [1, 4] is index 1*420 + 4*10 = 460.


@param radius (int)

else use neighborhood

github.com

numenta/nupic/blob/6924a904644bbb8cdba6ca7a62050cf8456614f8/src/nupic/math/topology.py#L81


42x10, the point [1, 4] is index 1*420 + 4*10 = 460.
"""
index = 0
for i, dimension in enumerate(dimensions):
  index *= dimension
  index += coordinates[i]


return index




def neighborhood(centerIndex, radius, dimensions):
"""
Get the points in the neighborhood of a point.


A point's neighborhood is the n-dimensional hypercube with sides ranging
[center - radius, center + radius], inclusive. For example, if there are two
dimensions and the radius is 3, the neighborhood is 6x6. Neighborhoods are
truncated when they are near an edge.


This is designed to be fast. In C++ it's fastest to iterate through neighbors
one by one, calculating them on-demand rather than creating a list of them.

topology.neighborhood(colIndex, sp._inhibitionRadius, sp._columnDimensions)

Topic		Replies	Views
Unecessary calls to sp._getColumnNeighborhood()? NuPIC	1	485	December 8, 2016
SP columns activating without input when using topology NuPIC spatial-pooling , topology	4	1183	August 23, 2016
Spatial Pooling Visualization NuPIC	6	1071	August 31, 2017
Frontiers Topology Paper Code Errors NuPIC newbie	9	842	April 8, 2019
Add method on SP to get initialization parameters NuPIC sequence-memory , proposal	3	789	July 30, 2016

How to replace call to sp._getNeighborsND()

Related Topics