next up previous contents
Next: Postprocessing Up: A Data Mining Tool Previous: Map visualization

Data on map


ENTIRE can be used to investigate local data sets. The local data sets are obtained by taking a set of input vectors and assigning each vector to the local data set of its BMU. The number of samples in each map unit gives the distribution of the data set on the map (see section 2.6). In ENTIRE the data histograms can be visualized by showing the number of samples in the set, or as a square the sidelength of which is proportional to the number of samples in the set, as seen in figures 4.5a and b.

ENTIRE can also be used to project arbitrary user-provided vectors on the map. Figure 4.6a shows the BMU search tool of ENTIRE and figure 4.6b shows the results of a search. The tool can be used to find multiple BMUs. Searching multiple BMUs visualizes well the ordering of the map and the familiarity of the sample vector: if the found BMUs are neighboring units, it can be assumed that the map is well organized. On the other hand BMUs far from each other mean a folding in the map topology: either the map is poorly trained, or the sample vector is atypical.

For time-series data analysis ENTIRE offers a trajectory tool shown in figure 4.7. The trajectory of a data set is formed of the consecutive BMUs of the data vectors in the data set. In ENTIRE the user can select the current vector, the length of trajectory following the current BMU, and the number of BMUs searched for the current vector.

Juha Vesanto
Tue May 27 12:40:37 EET DST 1997