• Post Reply
  • Bookmark Topic Watch Topic
  • New Topic

JavaFX with Kohonen SOM

 
Ben Taylor
Ranch Hand
Posts: 31
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
Hello,

I am using the Kohonen SOM Java from http://jknnl.sourceforge.net (code not maintained anymore) for mesh reconstruction and I am having rendering issues after the learning process is finished.

I am using a Hexagonal Matrix with a 50x50 grid, the learning rate is 0.8 and the algorithm "Winner Takes Most".

The issues is that after the learning process is finished which is supposed to "sort" the weights in the network, the mesh is always rendered incorrect.For example, the top point of the mesh is connected to the bottom with 2 other points to create a face.

I tried many settings but none render the mesh correctly and I thought there might be an issue with the method which returns the connected neurons.

I don't know where to go next, does anybody see an issue in the following method?

Thank you.


Screen Shot 2016-05-24 at 21.33.41.jpg
[Thumbnail for Screen Shot 2016-05-24 at 21.33.41.jpg]
Screen Shot 2016-05-24 at 21.29.22.jpg
[Thumbnail for Screen Shot 2016-05-24 at 21.29.22.jpg]
Screen Shot 2016-05-24 at 21.59.25.jpg
[Thumbnail for Screen Shot 2016-05-24 at 21.59.25.jpg]
 
Ben Taylor
Ranch Hand
Posts: 31
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
The picture market with 1 is the initial shape and the picture marked with 2 is the result. I think the method which returns the connected neurons is not bringing them back correctly.

Can somebody read this method better than me and see if there is an issue?
 
Ben Taylor
Ranch Hand
Posts: 31
  • Mark post as helpful
  • send pies
  • Quote
  • Report post to moderator
nobody ? please
 
  • Post Reply
  • Bookmark Topic Watch Topic
  • New Topic