| somQuality {aweSOM} | R Documentation |
SOM quality measures
Description
Computes several quality measures on a trained SOM (see Details).
Usage
somQuality(som, traindat)
Arguments
som |
|
traindat |
matrix containing the training data. |
Details
Four measures of SOM quality are returned :
- Quantization error:
Average squared distance between the data points and the map's prototypes to which they are mapped. Lower is better.
- Percentage of explained variance:
Similar to other clustering methods, the share of total variance that is explained by the clustering (equal to 1 minus the ratio of quantization error to total variance). Higher is better.
- Topographic error:
Measures how well the topographic structure of the data is preserved on the map. It is computed as the share of observations for which the best-matching node is not a neighbor of the second-best matching node on the map. Lower is better: 0 indicates excellent topographic representation (all best and second-best matching nodes are neighbors), 1 is the maximum error (best and second-best nodes are never neighbors).
- Kaski-Lagus error:
Combines aspects of the quantization and topographic error. It is the sum of the mean distance between points and their best-matching prototypes, and of the mean geodesic distance (pairwise prototype distances following the SOM grid) between the points and their second-best matching prototype.
Value
A list containing quality measures : quantization error, share
of explained variance, topographic error and Kaski-Lagus error (see
Details).
References
Kohonen T. (2001) Self-Organizing Maps, 3rd edition, Springer Press, Berlin. <doi:10.1007/978-3-642-56927-2>
Kaski, S. and Lagus, K. (1996) Comparing Self-Organizing Maps. In C. von der Malsburg, W. von Seelen, J. C. Vorbruggen, and B. Sendho (Eds.) Proceedings of ICANN96, International Conference on Articial Neural Networks , Lecture Notes in Computer Science vol. 1112, pp. 809-814. Springer, Berlin. <doi:10.1007/3-540-61510-5_136>