Distribution of the quantization error (QE, represented in different shades of green) and topographic error (TE, represented in different shades of purple) as a function of the map size of the SOM, assuming as neighborhood function a Gaussian (top panel) and a Mexican hat (bottom panel). In each plot, the number of training epochs is set to 200 (diamond markers), 1100 (circles) and 1700 (squares), respectively. The dotted line represents the optimal SOM size for representing our dataset with 1100 training epochs.