import numpy as np
import matplotlib.pyplot as plt
# W初始化
def w_init(row, col, data_rows):
w1 = np.zeros((row, col, data_rows)) # 构造W初始数组
for i in range(row):
for j in range(col):
w1[i][j] = np.random.rand(1, data_rows) # 为W随机赋值
return w1
# 数据预处理
def data_load():
f = open(r'C:\Users\lyb\Desktop\iris.txt') # 导入数据
data = []
data1 = []
for line in f.readlines():
data.append(line.strip().split(',')) # 切割TXT文件，构造数据列表
data.pop(0)
rows = len(data[0])-1 # 数据列表的行数
cols = len(data) # 数据列表的列数
for col in range(cols):
data1.append(data[col][0:rows])
data1 = np.array(data1, dtype=np.float32) # 转换格式：将列表字符格式转换为数组实数格式
for col in range(cols):
data2 = data1[col]/np.sqrt(sum(data1[col] ** 2))
data1[col] = data2
return data1, cols, rows
# W更新函数
def update_w(data, w, alpha_init, zeta_init, max_iter):
for iter1 in range(max_iter):
print('Calculating..................')
w_m, w_n, *p = np.shape(w) # 计算W的行数和列数
d = np.zeros((w_m, w_n), dtype=np.float32) # 构造距离d初始数组
t = np.random.randint(0, 150) # 随机选取一个输入数据
alpha = alpha_init * np.exp(- iter1/1000)
zeta = zeta_init * np.exp(- iter1/500)
for i in range(w_m):
for j in range(w_n):
d1 = np.sum((w[i][j] - data[t]) ** 2) # 计算输入数据与各神经元的距离
d[i][j] = np.sum(d1)
min_d = 2
for i in range(w_n):
for j in range(w_m):
if min_d >= d[i][j]:
min_d = d[i][j]
else:
min_d = min_d
ind1, ind2, *p = np.where(d == min_d) # 获取最优神经元的位置坐标
for i in range(w_n):
for j in range(w_m):
di = (i - ind1) ** 2 + (j - ind2) ** 2
h = np.exp(- (di / (2 * zeta ** 2)))
w[i][j] = w[i][j] + alpha * h * (data[t] - w[i][j]) # 更新各个权值
print('Done')
return w
# 数据可视化
def draw(data, w):
t1 = np.linspace(-0.5, 9.5, 11)
xx, yy = np.meshgrid(t1, t1)
w_m, w_n, *p = np.shape(w)
for t in range(cols):
d = np.zeros((w_n, w_m), dtype=np.float32)
for i in range(w_m):
for j in range(w_n):
d1 = np.sum((w[i][j] - data[t]) ** 2)
d[i][j] = np.sum(d1)
min_d = 2
for i in range(w_n):
for j in range(w_m):
if min_d >= d[i][j]:
min_d = d[i][j]
else:
min_d = min_d
ind1, ind2, *p = np.where(d == min_d)
plt.plot(t1, xx, 'w')
plt.plot(xx, t1, 'w') # 绘制网格
plt.title('Flower Cluster')
if t < 50:
plt.plot(ind1[0], ind2[0], 'r*')
if (50 < t) & (t < 100):
plt.plot(ind1[0], ind2[0], 'b*')
if (100 < t) & (t < 150):
plt.plot(ind1[0], ind2[0], 'g*')
plt.show()
if __name__ == '__main__':
data, cols, rows = data_load()
w = w_init(10, 10, rows)
w = update_w(data, w, 0.2, 3, 5000)
draw(data, w)
