SOM for Gerardo’s Data

1. Loading raw data and getting it ready

library(readr)
library(readr)
img <- read_csv("~/GENER/Gerardo/1. Raw Data/DatosDifsg.csv")
img <- img[,-1:-2]

The dataframe obtained is made of integer values:

str(img)

Classes ‘tbl_df’, ‘tbl’ and 'data.frame':   250 obs. of  6 variables:
 $ sgBaseDiffb: num  -9.94e-06 -7.44e-06 -3.55e-06 9.08e-07 5.14e-06 ...
 $ sg102Diffb : num  5.05e-06 3.19e-06 3.08e-06 3.99e-06 5.16e-06 ...
 $ sg113Diffb : num  1.73e-05 9.19e-06 3.74e-06 4.58e-07 -1.18e-06 ...
 $ sg130Diffb : num  3.62e-05 2.66e-05 1.79e-05 1.02e-05 3.60e-06 ...
 $ sg103Diffb : num  -1.69e-05 -1.57e-05 -1.17e-05 -6.23e-06 -3.03e-07 ...
 $ sg91Diffb  : num  2.58e-06 2.42e-06 1.81e-06 1.04e-06 4.16e-07 ...

This will be a problem down the line when feeding this data to the algorithm, values have to be changed to numeric:

#img[,'FILA'] <- as.numeric(as.factor(img$FILA))
#img[,'COLUMNA'] <- as.numeric(as.factor(img$COLUMNA))
#img[,'RED'] <- as.numeric(as.factor(img$RED))
#img[,'GREEN'] <- as.numeric(as.factor(img$GREEN))
#img[,'BLUE'] <- as.numeric(as.factor(img$BLUE))

Whit this taken care of we have a database ready to be fed to the algorithmL

#str(img)

2 Initial exploration of data

A brief summary of the values of the variable can help with the initial understanding of things:

summary(img)

  sgBaseDiffb           sg102Diffb           sg113Diffb           sg130Diffb           sg103Diffb        
 Min.   :-1.950e-05   Min.   :-1.997e-05   Min.   :-3.320e-05   Min.   :-3.283e-05   Min.   :-3.985e-05  
 1st Qu.:-5.326e-06   1st Qu.:-5.057e-06   1st Qu.:-5.765e-06   1st Qu.:-5.985e-06   1st Qu.:-7.367e-06  
 Median :-6.496e-07   Median : 6.249e-09   Median : 1.152e-07   Median :-3.131e-07   Median :-8.982e-07  
 Mean   : 3.176e-08   Mean   : 1.960e-08   Mean   : 2.278e-08   Mean   : 1.672e-07   Mean   :-3.965e-07  
 3rd Qu.: 5.720e-06   3rd Qu.: 4.881e-06   3rd Qu.: 5.511e-06   3rd Qu.: 6.917e-06   3rd Qu.: 6.313e-06  
 Max.   : 1.887e-05   Max.   : 2.203e-05   Max.   : 2.353e-05   Max.   : 3.616e-05   Max.   : 3.413e-05  
   sg91Diffb         
 Min.   :-4.790e-05  
 1st Qu.:-6.094e-06  
 Median :-3.885e-07  
 Mean   : 9.732e-08  
 3rd Qu.: 5.872e-06  
 Max.   : 4.821e-05  

3 Self Organizing Map

Loading required package

require(kohonen)

Changing dataframe to a matrix and normalizing it’s values

img.matrix <- as.matrix(img)
#TEST
#img.matrixTest <- img.matrix[1:1000000,]

Creating the SOM grid

som_grid <- somgrid(xdim = 6, ydim = 6, topo = "hexagonal")

Training the SOM

Training progress evaluation

plot(som_model, type = "changes")

Node counts

source('coolBlueHotRed.R')
plot(som_model, type = "counts", palette.name=coolBlueHotRed)

SOM by RGB value

source("coolBlueHotRed.R")
models <- as.data.frame(som_model$codes)
par(mfrow = c(3,3))
plot(som_model,
     type = "property",
     property = models[,1],
     main=names(models)[1],
     palette.name=coolBlueHotRed)
plot(som_model,
     type = "property",
     property = models[,2],
     main=names(models)[2],
     palette.name=coolBlueHotRed)

plot(som_model,
     type = "property",
     property = models[,3],
     main=names(models)[3],
     palette.name=coolBlueHotRed)
plot(som_model,
     type = "property",
     property = models[,4],
     main=names(models)[4],
     palette.name=coolBlueHotRed)

plot(som_model,
     type = "property",
     property = models[,5],
     main=names(models)[5],
     palette.name=coolBlueHotRed)
plot(som_model,
     type = "property",
     property = models[,6],
     main=names(models)[6],
     palette.name=coolBlueHotRed)

Neighbor Distance

plot(som_model, type = "dist.neighbours", palette.name = grey.colors)

Code Spread

plot(som_model, type = "codes", palette.name = coolBlueHotRed)

Clustering

mydata <- som_model$codes 
wss <- (nrow(mydata)-1)*sum(apply(data.frame(mydata),2,var)) 
for (i in 2:20) {
  wss[i] <- sum(kmeans(data.frame(mydata), centers=i)$withinss)
}
plot(1:20, wss, type="b", xlab="Number of Clusters",
     ylab="Within groups sum of squares", main="Within cluster sum of squares (WCSS)")

som_cluster <- cutree(hclust(dist(data.frame(som_model$codes))), 3)

In this graphic we can see that the ideal amount of clusters is not more than 5. Starting with 5 and going down we will choose the smallest number that gives us a smooth distribution in quantity of it’s members:

# 7 Clusters -- the first cluster has considerably less members than the other four ones.
table(cutree(hclust(dist(data.frame(som_model$codes))), 3))


# 5 Clusters -- the first cluster has considerably less members than the other four ones.
table(cutree(hclust(dist(data.frame(som_model$codes))), 5))

#4 clusters -- we have founde the good one.
table(cutree(hclust(dist(data.frame(som_model$codes))), 4))

pretty_palette <- c("#1f77b4", '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b', '#e377c2')
plot(som_model, type="mapping", bgcol = pretty_palette[som_cluster], main = "Clusters", keepMargins = TRUE) 
add.cluster.boundaries(som_model, som_cluster, lwd = 7)