def hor_connections(r1, x, y, stride, r2, w, d):
"""
Assume r1 is equal to r2
Given an (x,y) point in a square of size r2 x r2 find the
neurons which model the horizontal sides of a square of size
2 * stride + 1 in the square r1 x r1
In order to detect squares of different sizes, 5 sides are
computed, and the neurons on each of them have weights
determined by a gaussian distribution
x
+--------------------+
|
|
| ._________.
| stride
| (x,y)
|
| ._________.
+--------------------+
"""
out = []
for j, gw in enumerate(GAUSSIAN_WEIGHTS):
gaussians = list(
range(-(len(GAUSSIAN_WEIGHTS) // 2),
len(GAUSSIAN_WEIGHTS) // 2 + 1))
for i in range(-stride, stride + 1):
# Top side
pre_x = x + i
pre_y = y - stride + gaussians[j]
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
# Bottom side
pre_x = x + i
pre_y = y + stride + gaussians[j]
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
out = filter_connections(out)
return list(set(out))
def left_diag_connections(r1, x, y, stride, r2, w, d):
"""
Assume r1 is equal to r2
Given an (x,y) point in a square of size r2 x r2 find the
neurons which model the left diagonal \ sides of a square of size
2 * stride + 1 in the square r1 x r1
x
+--------------------+
|
|
| . (x,y)
| \
| \
| .
|
+--------------------+
"""
out = []
for j, gw in enumerate(GAUSSIAN_WEIGHTS):
gaussians = list(
range(-(len(GAUSSIAN_WEIGHTS) // 2),
len(GAUSSIAN_WEIGHTS) // 2 + 1))
for i in range(0, 2 * stride + 1):
# Left side
pre_x = x - 2 * stride + i + gaussians[j]
pre_y = y + i - gaussians[j]
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
# Right side
pre_x = x + i + gaussians[j]
pre_y = y - 2 * stride + i - gaussians[j]
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
out = filter_connections(out)
return list(set(out))
def vert_connections(r1, x, y, stride, r2, w, d):
"""
Assume r1 is equal to r2
Given an (x,y) point in a square of size r2 x r2 find the
neurons which model the vertical sides of a square of size
2 * stride + 1 in the square r1 x r1
x
+--------------------+
|
|
| .stride .
| | |
| | (x,y) |
| | |
| . .
+--------------------+
"""
out = []
for j, gw in enumerate(GAUSSIAN_WEIGHTS):
gaussians = list(
range(-(len(GAUSSIAN_WEIGHTS) // 2),
len(GAUSSIAN_WEIGHTS) // 2 + 1))
for i in range(-stride, stride + 1):
# Left side
pre_x = x - stride + gaussians[j]
pre_y = y + i
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
# Right side
pre_x = x + stride + gaussians[j]
pre_y = y + i
out.append((neuron_id(pre_x, pre_y, r1), neuron_id(x, y,
r2), w * gw, d))
out = filter_connections(out)
return list(set(out))
def vertical_connectivity_pos(res1, x, y, res2):
"""
Return the positive connections of the following
receptive field
x
-2 -1 0 1 2
-2 - -
-1 - + -
y 0 - + -
1 - + -
2 - -
"""
out = []
x2 = x / (res1 / res2)
y2 = y / (res1 / res2)
# Positive
out.append((neuron_id(x , y-1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x , y , res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x , y+1, res1), neuron_id(x2 , y2, res2)))
out = filter_connections(out)
return out
def horizontal_connectivity_pos(res1, x, y, res2):
"""
Return the positive connections of the following
receptive field
x
-2 -1 0 1 2
-2
-1 - - - - -
y 0 + + +
1 - - - - -
2
"""
out = []
x2 = x / (res1 / res2)
y2 = y / (res1 / res2)
# Positive
out.append((neuron_id(x-1, y , res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x , y , res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x+1, y , res1), neuron_id(x2 , y2, res2)))
out = filter_connections(out)
return out
def split_pos_neg_spikes(self):
out_pos = []
out_neg = []
n_neurons = self.cam_res * self.cam_res
for _ in range(n_neurons):
out_pos.append(list())
out_neg.append(list())
for spikes_time in self.output_spikes_tuple:
# Split tuple
spike = spikes_time[0]
t = spikes_time[1]
x = spike[0]
y = spike[1]
polarity = spike[2]
if polarity:
out_pos[neuron_id(x, y, self.cam_res)].append(t)
else:
out_neg[neuron_id(x, y, self.cam_res)].append(t)
return out_pos, out_neg
def right_diagonal_connectivity_neg(res1, x, y, res2):
"""
Return the negative connections of the following
receptive field
x
-2 -1 0 1 2
-2 -
-1 - - + -
y 0 - + -
1 - + - -
2 -
"""
out = []
x2 = x / (res1 / res2)
y2 = y / (res1 / res2)
# Negative
out.append((neuron_id(x-2, y+1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x-1, y , res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x-1, y-1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x , y-1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x+1, y-2, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x-1, y+2, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x , y+1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x+1, y+1, res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x+1, y , res1), neuron_id(x2 , y2, res2)))
out.append((neuron_id(x+2, y-1, res1), neuron_id(x2 , y2, res2)))
out = filter_connections(out)
return out
