Comments on: The Self-Organized Gene (Part 2) http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/ The Peltarion Blog Tue, 22 Dec 2009 17:29:17 +0000 http://wordpress.org/?v=2.9.1 hourly 1 By: Luka (Peltarion) http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-24553 Luka (Peltarion) Mon, 03 Nov 2008 10:31:21 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-24553 The reason why the U-matrix doesn't look like the clusters is because it hasn't been used. U-matrices can be difficult to interpret and there are better ways of finding clusters. The actual algorithm used is a neural gas that finds the clusters in the SOM. The reason why the U-matrix doesn’t look like the clusters is because it hasn’t been used. U-matrices can be difficult to interpret and there are better ways of finding clusters. The actual algorithm used is a neural gas that finds the clusters in the SOM.

]]> By: JackK http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-24457 JackK Thu, 30 Oct 2008 18:34:29 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-24457 Ok. Next question (the last, I think). The image titled "Clusters" does not look like (for me) to be created according to that "10-dimension" U-matrix. The U-matrix for "Column0" clearly shows that there are 3 clusters. However from "10-dimension" U-matrix it's hard (for me) to see the areas of 3 clusters you showed. Can you help me again? Ok.
Next question (the last, I think). The image titled “Clusters” does not look like (for me) to be created according to that “10-dimension” U-matrix. The U-matrix for “Column0″ clearly shows that there are 3 clusters. However from “10-dimension” U-matrix it’s hard (for me) to see the areas of 3 clusters you showed. Can you help me again?

]]> By: Luka http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-24455 Luka Thu, 30 Oct 2008 18:01:52 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-24455 The U-matrix is calculated as the distance between the nodes. It doesn't matter in how many dimensions the map itself is - it's still standard euclidian distance that is used. So for each node the distance to its neighbours is calculated. The U-matrix is calculated as the distance between the nodes. It doesn’t matter in how many dimensions the map itself is – it’s still standard euclidian distance that is used. So for each node the distance to its neighbours is calculated.

]]> By: JackK http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-24453 JackK Thu, 30 Oct 2008 16:13:40 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-24453 Hi, I understand maplets very well. I also understand U-matrix from your example (created from image of feature named "Column0"). I can't, however, understand the U-matrix which is on the image below the sentence "Now you should understand how to read and interpret a SOM displayed like this one in Synapse:". There is only 1 U-matrix but we have 10 features (from Population" to "Cows"). That is that U-matrix's construction? Please explain me this in simple words (my english is not very good). Thanks in advance. Hi,

I understand maplets very well. I also understand U-matrix from your example (created from image of feature named “Column0″). I can’t, however, understand the U-matrix which is on the image below the sentence “Now you should understand how to read and interpret a SOM displayed like this one in Synapse:”. There is only 1 U-matrix but we have 10 features (from Population” to “Cows”). That is that U-matrix’s construction?
Please explain me this in simple words (my english is not very good).
Thanks in advance.

]]> By: Self-organizing map http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-12511 Self-organizing map Thu, 16 Aug 2007 07:48:33 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-12511 [...] “The Self-Organized Gene, Part 1″, and Part 2 beginners’ level introduction to competitive learning and self-organizing maps. [...] [...] “The Self-Organized Gene, Part 1″, and Part 2 beginners’ level introduction to competitive learning and self-organizing maps. [...]

]]> By: Luka (Peltarion) http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-11306 Luka (Peltarion) Wed, 25 Jul 2007 06:15:21 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-11306 Hi, For questions about Synapse, send an email to info(AT)peltarion.com. It's better that we keep the blog comments on topic. Just send an email and we'll be more than happy to answer your questions. -Luka / Peltarion Hi,

For questions about Synapse, send an email to info(AT)peltarion.com. It’s better that we keep the blog comments on topic.

Just send an email and we’ll be more than happy to answer your questions.

-Luka / Peltarion

]]> By: fatty http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-11291 fatty Wed, 25 Jul 2007 00:54:07 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-11291 hi, another question from me.. 1. Why do the sensisivity of the inputs are not constant to the output. If i pause/stop/reset and train the netowrk again, it will show a different level of sensitivity. I thought the sensitivity should remain the same regardless the no. of training? thanks. hi,

another question from me..

1. Why do the sensisivity of the inputs are not constant to the output. If i pause/stop/reset and train the netowrk again, it will show a different level of sensitivity. I thought the sensitivity should remain the same regardless the no. of training?

thanks.

]]> By: fatty http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-11285 fatty Tue, 24 Jul 2007 21:35:34 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-11285 Hi again, 1. I couldn't find any section where i can set the stopping critera..I only found one where i can set the no. of epoch, but i dont find any on the error part. I want the training to stop once the error meets 0. 2. Also, in the pre-processing, i only set the percentage for validation. May I know how is the test sample being calculated and when is the test sample being used? Thanks!! Hi again,

1. I couldn’t find any section where i can set the stopping critera..I only found one where i can set the no. of epoch, but i dont find any on the error part. I want the training to stop once the error meets 0.

2. Also, in the pre-processing, i only set the percentage for validation. May I know how is the test sample being calculated and when is the test sample being used?

Thanks!!

]]> By: fatty http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-11232 fatty Mon, 23 Jul 2007 22:54:31 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-11232 Thanks a lot for the answers! Thanks a lot for the answers!

]]> By: Luka (Peltarion) http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/comment-page-1/#comment-11219 Luka (Peltarion) Mon, 23 Jul 2007 15:59:41 +0000 http://blog.peltarion.com/2007/06/13/the-self-organized-gene-part-2/#comment-11219 Hi, 1) The names don't have any functional effect - they're just labels. Unless you specify a name of your own, they are numbered automatically in the order they are laid out the first time. 2) http://www.peltarion.com/WebDoc/Components/index.html 3) You can't. Since a component can be basically anything, there is no guarantee at all that it has a simple weight structure. You can however deploy the system and use it outside of Synapse. --Luka / Peltarion Hi,

1) The names don’t have any functional effect – they’re just labels. Unless you specify a name of your own, they are numbered automatically in the order they are laid out the first time.

2) http://www.peltarion.com/WebDoc/Components/index.html

3) You can’t. Since a component can be basically anything, there is no guarantee at all that it has a simple weight structure. You can however deploy the system and use it outside of Synapse.

–Luka / Peltarion

]]>