x[0][0] = pettern0[0]
x[0][1] = pettern0[1]
x[0][2] = pettern0[2]
x[0][3] = pettern0[3]
wij = np.zeros((100, 100))
for i in range(100):
for j in range(100):
for p in range(4):
wij[i][j] += (2 * x[0][p][i] - 1) * (2 * x[0][p][j] - 1)
wij[i][j] = wij[i][j] / 4
process_bar = lab.ShowProcess(t, "ok")
for t in range(t):
for n in range(4):
for i in range(100):
wij_some = 0
for j in range(100):
wij_some += wij[i][j] * x[t][n][j]
ni[t + 1][n][i] = kf * ni[t][n][i] + wij_some
ci[t + 1][n][i] = kr * ci[t][n][i] - ar * x[t][n][i] + a0
x[t + 1][n][i] = lab.sigmoid(ni[t + 1][n][i] + ci[t + 1][n][i])
lab.number_to_image(x[t][0], t)
process_bar.show_process()
"""
lab.number_to_image(x[0][0],10001)
lab.number_to_image(x[0][1],10002)
lab.number_to_image(x[0][2],10003)
lab.number_to_image(x[0][3],10004)
"""
for s in range(100):
max_ci += abs(ni[t][n][s])
for j in range(100):
wij_some += wij[i][j]*x[t][n][j]
max_ci = max_ci/M
ni[t+1][n][i] = kf*ni[t][n][i] + wij_some
ci[t+1][n][i] = kr*ci[t][n][i] - ar*x[t][n][i] + a0
if ci[t+1][n][i] >= max_ci and t != 0:
ci[t+1][n][i] = max_ci
elif ci[t+1][n][i] <= -max_ci and t != 0:
ci[t+1][n][i] = -max_ci
x[t+1][n][i] = lab.sigmoid(ni[t+1][n][i]+ci[t+1][n][i])
lab.number_to_image(x[t][pattern],t)
process_bar.show_process()
"""
for i in range(4):
lab.number_to_image(x[0][i], 1000 + i)
for i in range(4):
sum = 0
for j in range(100):
sum += x[0][i][j]
print(sum)
print(x[0][0])
for i in range(100):
for j in range(100):
for p in range(4):
wij[i][j] += (2 * x[0][p][i] - 1) * (2 * x[0][p][j] - 1)
wij[i][j] = wij[i][j] / 4
for t in range(t):
for n in range(4):
for i in range(100):
wij_some = 0
max_ci = 0
for s in range(100):
max_ci += abs(ni[t][n][s])
for j in range(100):
wij_some += wij[i][j] * x[t][n][j]
max_ci = max_ci / M
ni[t + 1][n][i] = kf * ni[t][n][i] + wij_some
ci[t + 1][n][i] = kr * ci[t][n][i] - ar * x[t][n][i] + a0
if ci[t + 1][n][i] >= max_ci and t != 0:
ci[t + 1][n][i] = max_ci
elif ci[t + 1][n][i] <= -max_ci and t != 0:
ci[t + 1][n][i] = -max_ci
x[t + 1][n][i] = lab.sigmoid(ni[t + 1][n][i] + ci[t + 1][n][i])
lab.number_to_image(x[t][pattern], t)
process_bar.show_process()
for s in range(100):
if t != 0:
u_sum += abs(x[t][partten][s]-es*x[t-1][partten][s])
for j in range(100):
wij_some += wij[i][j]*x[t][n][j]
ni[t+1][n][i] = kf*ni[t][n][i] + wij_some
ci[t+1][n][i] = kr*ci[t][n][i] - ar*(belt**(K*u_sum))*x[t][n][i] + a0
x[t+1][n][i] = lab.sigmoid(ni[t+1][n][i]+ci[t+1][n][i])
process_bar.show_process()
#lab.number_to_image(x[t][partten],t)
plt.scatter(t,u_sum,s = 0.1, c = 'b')
plt.show()
"""
lab.number_to_image(x[0][0],1)
lab.number_to_image(x[0][1],2)
lab.number_to_image(x[0][2],3)
lab.number_to_image(x[0][3],4)
"""
"""
if we change from u_sum += abs(x[t][partten][s]-x[t-1][partten][s])
to u_sum += abs(x[t][partten][s]-x[t-1][partten][s]/2)
it's obvious that we get a higher quality of control
"""