A large number of practical problems involves elements that are
described as a mixture of qualitative and quantitative information,
and whose description is probably incomplete. The
self-organizing map is an effective tool for visualization of
high-dimensional continuous data. In this work, we extend the network
and training algorithm to cope with heterogeneous information, as well
as missing values. The classification performance on a collection of
benchmarking data sets is compared in different configurations.
Various visualization methods are suggested to aid users interpret
post-training results.