Self organizing map and Kohonen's in knime

KNIME Analytics Platform
#1

Dear  all,

 

Is it possible to have in the mining tree a node for Self organizing map and kohonen's map?

 

thanks a lot

#2

Dear all,

 

I found in the weka neural network web site http://jsalatas.ictpro.gr/weka/ develop SOM algorithms ...is it possible to add to weka? or can anyone tell me how to do easily? i think that SOM could be a good improvment in knime dataming capabilities

 

thanks in advance

#3

For weka 3.7.2 or above the SOM algorithm can be added automatically through weka's package managememnt system.

See http://weka.wikispaces.com/How+do+I+use+the+package+manager%3F for details.

#4

Any pointers on how to make KNIME use Weka 3.7.2? The newest I've found in my KNIME installation is 3.7.1beta, apparently supplied with the HCS tools and probably not called by default.

FWIW, on my end the interest is to integrate the MARSplines algorith featured at http://cns.bu.edu/~gsc/CN710/pmwiki.php?n=Main.MARSplines, kindly provided to me by the original author. Not even sure whether the package manager is the right tool for this purpose. Anyway, I'm hoping that a Weka implementation won't suffer from the same memory restrictions as the two R implementations of MARS do (such as the otherwise excellent "earth" package, owed to general R limits). It's truly a shame not to have decent mining capabilities for continuous responses on bigger data sets... :-(

Thanks
E

#5

dear john,

Thanks ofr informations , my problem is as in the newest post how to have this and other new algorithms in weka node in knime. Is there any way how to do that? in knime weka version is 3.7.1 beta not 3.7.2 can be updated with this alg extensions? or we need to wait untill v2.5 of knime is ready?

 

thanks

I think that could be usefull to have in weka this chance to update alg without build a node.

 

regards

#6

Maybe you can try an R function via the R nodes.

R nodes communicates with an R installation that is supported via the Knime R plugin.

Within the R node you have to import the package you want to use and call the function with your data. Sometimes you even have to install additional packages to your R implementation.

For example:

http://cran.r-project.org/web/packages/som/index.html

There are several examples in knime and in the labs which show how to use the R nodes.

When you are fit in programming that is also a nice idea to create new nodes. Just wrap the R call, create a nice knime user interface and hide the R call to the user. I once did that for the local outlier factor alg.

If you are interested and have problems somewhere let me know.

Cheers

#7

thanks

I'm not use very often R, I have install the som and kohonen package in R, i tried to use som function with the iris data example , i used

library(kohonen)
R<-som(class~.,R)

in r model node, but this error

Error in som(class ~ ., R) : Argument data should be numeric

Do you have some example how to use this and visualize the map? correctly

 

thanks a lot

#8

There is an example here:

http://cran.r-project.org/web/packages/kohonen/kohonen.pdf

In the example the map is trained first with the training data and then the prediction is calculated. In the end a table with both is combined.

There are several examples in the labs and the forum how you add table rows and colums as arguments for R functions.

http://www.knime.org/introduction/examples  (R example)

http://tech.knime.org/forum/r-statistics-nodes-and-integration

#9

dear al

Thanks for your indications , i have imported in R the package som and kohonen.

I tried to use a snippet node writing this code

library("som")
R<-data(R$"[1:4])
R<-iris.sc <- scale(R$"[1:4]")
R<-som <- som(data = iris.sc, xdim = 5, ydim = 5)
R<-plot(som)

but this dosen't work well, the error outpit is that it try to write a csv file?

Error: unexpected string constant in:
"R<-plot(som)
write.csv(R, ""
Execution halted

 

can you help me ?
 

#10

hm no idea..  maybe if you want to plot sth. you should use the R view node?

 where is your training and your prediction?

what is R<-som <- som(data = iris.sc, xdim = 5, ydim = 5)?

 

take a look at the code in the pdf (page 14)