0

I want to count quantization error (qe) in my SOM model using R. This is the code I've tried:

data(yeast)## take only complete 
x <- yeast[[3]][apply(yeast[[3]], 1, function(x) sum(is.na(x))) == 0,]
yeast.som <- som(x, somgrid(5, 8, "hexagonal"))
## quantization error:
mean(yeast.som$distances)

Is that code right? There exist another method to count qe?

Andrea Ianni
  • 829
  • 12
  • 24
Hanima
  • 25
  • 7

1 Answers1

1

It is right to use yeast.som$distances to give an errore measure related to a SOM because it represents the mean distance of the elements from their codebooks. If I had to choose only one feature I would choose that one.

Nevertheless that is not the only one: for example you could compute the number of empty cells (cells that have no elements assigned) starting from yeast.som$unit.classif. The lower number of empty cells, the better SOM.

You have also another measure about the distance of each codebook from the other ones. You can see it doing: plot(yeast.som, type = "quality").

A measure of goodness of a SOM could be a combination of the one presented above.

Andrea Ianni
  • 829
  • 12
  • 24