We can create models as we like, but before a model can be reliably used, it must be validated. Validation means that the model is tested so that we can be sure that the model gives us reasonable and accurate values. What we mean by this depends largely on the application and our requirements.

The validation must be done by using an independent test set. The independent test set is a set similar to the input set but not a part of the training set. The testing set can be seen as a representative of the general case.

Quantization error of an input vector is defined as the difference between the input vector and the closest codebook vector. For a set of input vectors, we can reflect on the similarity of the input data set and the SOM by investigating the distribution of the quantization errors. The range of quantization error tells the smallest and the largest amount of error.

Fri Mar 8 13:44:32 EET 1996