Add self-organizing maps or generative topographical maps as a scenario discovery technique

[quaquel]

See Bonham et al

Basic idea is to train the map on the full experiment + outputs, distance calculation in the input space is straightforward but in the output space can be a bit tricky. Next, you can create slices for both inputs and outputs. For outputs, you can see where types of outcomes land in the map, while on the input side, you can then see what the common denominator is of the cluster of outcomes.