
Fig. 5.6 Quality parameters for SOMs: the plot on the left shows the decrease in distance between objects and their closest codebook vectors during training. The plot on the right shows the mean distances between objects and codebook vectors per unit This leads to the plots in Fig. 5.6. The left plot shows the average distance (expressed per variable) to the winning unit during the training iterations, and the right plot shows the average distance of the samples and their corresponding codebook vectors after training. Note that the latter plot concentrates on distances within the unit whereas the Umatrix plot in Fig. 5.5 visualizes average distances between neighboring units. Finally, an indication of the quality of the map is given by the mean distances of objects to their units: > summary(wines.som) SOM of size 5x4 with a hexagonal topology and a bubble neighbourhood function.
The number of data layers is 1. Distance measure(s) used: sumofsquares. Training data included: 177 objects. Mean distance to the closest unit in the map: 3.646. The summary function indicates that an object, on average, has a distance of 3.6 units to its closest codebook vector 3xFLAG formula. The plot on the left in Fig. 5.6 shows that the average distance drops during training: codebook vectors become more similar to the units that are mapped to them. The plot on the right, finally, shows that the distances within units can be quite different Lipo3000 Transfection Reagent. Interestingly, some of the units with the largest spread only contain Grignolinos (units 2 and 8), so the variation can not be attributed to class overlap alone.

