I have a question about output of my SOM network. I have trained my network with diffrent size, learning rate and epochs, but my output always can recognise two big clusters. Iris-setosa and Iris-virginica, both classes has fitted well in two BMUs. But output of Iris-versicolor is very different. It contain some other classes BMU but this is not a big problem.
I just want to know is this good or I have some type of bug?

Setosa
0. 1846
1. 1846
2. 1846
3. 1846
4. 1846
5. 1846
6. 1846
7. 1846
8. 1846
9. 1846
10. 1846
11. 1846
12. 1846
13. 1846
14. 1846
15. 1846
16. 1846
17. 1846
18. 1846
19. 1846
20. 1846
21. 1846
22. 1846
23. 1846
24. 1846
25. 1846
26. 1846
27. 1846
28. 1846
29. 1846
30. 1846
31. 1846
32. 1846
33. 1846
34. 1846
35. 1846
36. 1846
37. 1846
38. 1846
39. 1846
40. 1846
41. 1620
42. 1846
43. 1846
44. 1846
45. 1846
46. 1846
47. 1846
48. 1846
49. 1846

Versicolor
50. 652
51. 652
52. 652
53. 1259
54. 696
55. 1394
56. 652
57. 490
58. 696
59. 490
60. 490
61. 1059
62. 1304
63. 696
64. 490
65. 652
66. 1400
67. 490
68. 696
69. 490
70. 652
71. 1574
72. 696
73. 832
74. 696
75. 696
76. 696
77. 652
78. 696
79. 490
80. 490
81. 490
82. 444
83. 696
84. 1129
85. 1084
86. 652
87. 696
88. 25
89. 584
90. 490
91. 789
92. 1034
93. 490
94. 854
95. 29
96. 584
97. 877
98. 490
99. 809

Virginica
100. 652
101. 696
102. 652
103. 652
104. 652
105. 652
106. 877
107. 652
108. 696
109. 652
110. 652
111. 696
112. 652
113. 696
114. 652
115. 652
116. 652
117. 652
118. 652
119. 696
120. 652
121. 652
122. 652
123. 696
124. 652
125. 652
126. 696
127. 652
128. 652
129. 652
130. 652
131. 652
132. 652
133. 696
134. 696
135. 652
136. 652
137. 652
138. 652
139. 652
140. 652
141. 652
142. 696
143. 652
144. 652
145. 652
146. 696
147. 652
148. 652
149. 652

Thanks in advance. ok now i have diagram it looks not bad. enter image description here

  • I have edited the question to a more readable format..please check whether the meaning has remain unaltered or not. – DuttaA Jun 14 at 2:08
  • Thank you so much! sorry for my english. Only one thing change meaning. on all the 3 flower datase - i want to say diffrent size of network, learning rate, epochs – Ameba kupiec Jun 14 at 2:12
  • I think you should add it to the question..i'll check the grammatical errors and make it readable – DuttaA Jun 14 at 2:14
  • I think you should add a scatter plot to represent the actual points vs the area covered by som – DuttaA Jun 15 at 8:46

This is not a bug, assuming you hve implemented the SOM properly with best decay rules of learning rate and neighbourhood strength (in short best hyperparameters).

Think of the map as finding many local clusters in the Iris-Versicolor class. Since SOM's behave somewhat like k-means clustering, I would say this implies Iris Versicolor has many local clusters which are as closely knit together as the entire classes of Iris setosa and Iris virginica.

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.