Clarify that multi-class example isn't expected to yield predictions

[ethanwhite]

Our current multi-class example in the docs trains on a single image for 1 epoch. It unsurprisingly doesn't learn anything and therefore predicts zero trees. This is confusing - see #773. We should clarify this in the docs and perhaps provide an idea of how many epochs the example would have to be trained for to show any results (~500 and then it only predicts dead trees).