How to relate back to original data points in a self organizing map

Question

I am using R kohonen package for the implementation of SOM. I find trouble in relating the code vector resulted from the self organizing map back to the original data points. I tried to include labels with no weight during the training process, but the result was incomprehensible.

Is there a way to refer back to the original data points from each node after the training process is complete?

Answer 1

You will get the center and scaled values from

x= attr(som_model$data,"scaled:center")

y= attr(som_model$data,"scaled:scale")

To get original data back

First find the node

som_model$unit.classif will return wining nodes corresponding to total number of observations.

Suppose you want to find out data related to the nth node then,

for (i in 1:ncol(som_model$data)){
 z[,i] = som_model$data[,i][som_model$unit.classif==n] * y[i]+x[i]
}

Corresponding to nth node you will get your original value back.