# BIC F20 Assignment 2 Problem 1
# Written by Tristan Kilper
import numpy as np
# Sample feature set + labels
feature_set = np.array([[66.5,66.5],[67,66.5],[66.5,67],[67,67]])
labels1 = np.array([66.5,66.5,66.5,67])
labels1 = labels1.reshape(4,1)
# random seed
np.random.seed(42)
# input neuron x
def in_x(Ix, t):
Cm = 1.6
Rm = 1.5
input_I = Ix
Vm = 2
thold = 0
Vr = -65
Vx = 66
def out_V(Vcur, t):
return ((-1 * (Vcur - Vr) + input_I*t)/Cm) - (Vm*t)/(Rm*Cm)
time = t # ms
cur_t = 0
cur_out = Vr
while cur_t <= time:
print('t: ' + str(cur_t) + ' V: ' + str(cur_out))
cur_out += out_V(cur_out, 1)
if cur_out >= thold:
print('[spike!]')
cur_out = Vr
Vx += .5
cur_t += 1
return Vx
# input neuron y
def in_y(Iy, t):
Cm = 1.6
Rm = 1.5
input_I = Iy
Vm = 2
thold = 0
Vr = -65
Vy = 66
def out_V(Vcur, t):
return ((-1 * (Vcur - Vr) + input_I*t)/Cm) - (Vm*t)/(Rm*Cm)
time = t # ms
cur_t = 0
cur_out = Vr
while cur_t <= time:
print('t: ' + str(cur_t) + ' V: ' + str(cur_out))
cur_out += out_V(cur_out, 1)
if cur_out >= thold:
print('[spike!]')
cur_out = Vr
Vy += .5
cur_t += 1
return Vy
# output neuron z
def out_z(Iz, t):
Cm = 1.6
Rm = 1.5
input_I = Iz
Vm = 2
thold = 0
Vr = -65
Vz = 66
def out_V(Vcur, t):
return ((-1 * (Vcur - Vr) + input_I*t)/Cm) - (Vm*t)/(Rm*Cm)
time = t # ms
cur_t = 0
cur_out = Vr
while cur_t <= time:
print('t: ' + str(cur_t) + ' V: ' + str(cur_out))
cur_out += out_V(cur_out, 1)
if cur_out >= thold:
print('[spike!]')
cur_out = Vr
Vz += .5
cur_t += 1
return Vz
# learning rules
a1 = 0.0000005
a2 = 0.0000001
a3 = 0.0000001
def d_w1(x,z):
return (a1*x*z) - (a2*x) - (a3*z)
b1 = 0.0000005
b2 = 0.0000001
b3 = 0.0000001
def d_w2(y,z):
return (b1*y*z) - (b2*y) - (b3*z)
# translate spike train to inputs
def readSpikeTrain(featset):
x_ins = np.array([-1,-1,-1,-1])
y_ins = np.array([-1,-1,-1,-1])
for n in range(4):
x_ins[n] = in_x(featset[n,0], 10)
y_ins[n] = in_y(featset[n,1], 10)
# x_ins = x_ins.reshape(4,1)
# y_ins = y_ins.reshape(4,1)
inputs = np.array([x_ins, y_ins])
inputs = inputs.T
return inputs
# train network
def train_network(inputs, labels):
weights = np.random.rand(2,1)
weights /= 2.55
print(inputs)
for epochs in range(100):
print(weights)
# output value
Vz_hat = []
Iz_hat = np.dot(inputs, weights)
print(Iz_hat)
for I in Iz_hat:
Vz_hat.append(out_z(I, 10))
print(Vz_hat)
for num in range(4):
weights[0] += d_w1(inputs[num,0], Vz_hat[num])
weights[1] += d_w2(inputs[num,1], Vz_hat[num])
if weights[0] > .6125 or weights[1] > .6125:
break
if weights[0] > .6125 or weights[1] > .6125:
print(weights)
break
return weights
test_inputs = readSpikeTrain(feature_set)
fin_weights = train_network(test_inputs, labels1)
# final test for all cases
single_point1 = np.array([66.5, 67])
single_point2 = np.array([66.5, 66.5])
single_point3 = np.array([67, 66.5])
single_point4 = np.array([67, 67])
result1 = out_z(np.dot(single_point1, fin_weights), 10)
result2 = out_z(np.dot(single_point2, fin_weights), 10)
result3 = out_z(np.dot(single_point3, fin_weights), 10)
result4 = out_z(np.dot(single_point4, fin_weights), 10)
print("Final Weights:")
print(fin_weights)
print("Final Test Cases:")
print(single_point1)
print(result1)
print(single_point2)
print(result2)
print(single_point3)
print(result3)
print(single_point4)
print(result4)
