C#/.net>> kohonen-network-with-wine-dataset>> 返回
项目作者： p208p2002

项目描述 ：
Unsupervised clustering for the UCI-WINE dataset using Kohonen Network
高级语言： Python
项目地址: git://github.com/p208p2002/kohonen-network-with-wine-dataset.git


Kohonen Network with Wine Dataset

Dataset

Kohonen Network

Kohnen Network使用@alexarnimueller所撰寫的Source Code

Source Code Ref: Kohonen, T. Self-Organized Formation of Topologically Correct Feature Maps. Biol. Cybern. 1982, 43 (1), 59–69.

Feature Normalization

進行資料特徵歸一化，為了更佳的分群效果

  1. def feature_normalization(X):
  2.     count_featrue = X.shape[-1]
  3.     f_maxs = []
  4.     f_mins = []
  6.     for i in range(count_featrue):
  7.         f_max = np.max(X[:,[i]])
  8.         f_min = np.min(X[:,[i]])
  9.         f_maxs.append(f_max)
  10.         f_mins.append(f_min)
  12.     for i in range(len(X)):
  13.         x_normalizatoion = np.array([0.0]*13)
  14.         for j in range(len(X[i])):
  15.             f_max = f_maxs[j]
  16.             f_min = f_mins[j]
  17.             x_normalizatoion[j] = (X[i][j] - f_min)/(f_max - f_min)
  18.         X[i] = x_normalizatoion
  19.     return X

Kohonen Network Training

  • network size : 8*8
  • epoch : 10000
  • learning rate : lr = 1 / (1 + (epoch / 0.5) **4)

Results

網路權重熱力圖

各別類別權重熱力圖







分群結果視覺化