It would be nice to have an example of showing train vs. test loss during training to understand over/under-fitting. The current setup doesn't make it obvious how to do this, and basically everyone using the experiment training interfaces is likely to do want to do this at some point.
Per this interesting question from Kiran: https://groups.google.com/forum/#!topic/clojure-cortex/8sLokBNmT2c
It would be nice to have an example of showing train vs. test loss during training to understand over/under-fitting. The current setup doesn't make it obvious how to do this, and basically everyone using the experiment training interfaces is likely to do want to do this at some point.