Any thoughts on which underlying matrix implementation works best for core.matrix? When I’ve used core.matrix I used clatrix because it had support for the features I needed. I personally was leaning towards nd4clj because I wanted to interop with dl4j more nicely. Now the cortex ecosystem is catching my eye. core.matrix is awesome, but the underlying matrix implementations still seem fractured…

cortex uses core.matrix. I believe the most fully implemented 'impl' for core.matrix is vectorz, but it works with several others (including clatrix as you mention). There is some work on having neanderthal be an impl. I don't think it is correct to think of the impls as being 'fractured' at all. One of the neat things about core.matrix is it can have many different impls, and they can even be used simultaneously and interoperably. The 'downside' of that is if you aren't paying attention you can get 'unexpected' performance hits.