Quantify similarity between self-organizing maps (SOMs)

What would be a valid similarity measure to quantify the (dis)similarity between two different datasets processed using the same trained version of a self-organizing map (trained on the combination of both datsets)?

What I mean is, each instance from the first dataset will be assigned to a specific node, giving rise to representation $R1$. Every instance from dataset 2 will also be assigned to a specific node, giving rise to representation $R2$. How do I quantify the dissimilarity between $R1$ and $R2$?

Would similarity measures for count data be appropriate, counting the number of datapoints assigned to a certain node of the map?

• Are you training the map on both datasets? In other words, is your training set just a combination of the two individual datasets? – KirkD_CO Jan 20 '18 at 16:38
• That's indeed the case – Archie Jan 20 '18 at 16:48