So the MNIST dataset is now included in my dissertation as a reference to see how well the SOM and other clustering methods are working.
![](https://cdn.sanity.io/images/qf8cbyxs/production/2eefea5cc6c93ffa16e0187cb68161f4d704828f-1026x760.png?w=1026&h=760&auto=format)
It looks a little bit squiffy, but I'm sure once the topographic error is calculated and a perfect lattice size is given it'll smooth out (hopefully).