unpack-feature-vector could be cleaner

[robinkraft]

See discussion.

Aaron suggests modifying the function so we can replace the current mess with something akin to this:

(map thrift/unpack seq)

As structured now, that wouldn't work in this function since we're only taking bits and pieces of the various inputs to create the feature vector, and the feature vector has a very specific ordering.

[robinkraft]

Could we use a feature vector map to make the feature vector a little cleaner? It's not cool counting on the position of values to ensure that the feature vector is created correctly. It'd be something like

{:min-break -1 :avg-break -0.5 ,,,,}

Instead of knowing that the last two elements in the feature vector are reserved for the hansen stat, and that min always comes before avg in the feature vector (I think).

[robinkraft]

Of course then it wouldn't be a vector anymore, but we could dump the values into a vector just before estimation. The order of elements doesn't matter as far as the statistical model is concerned. And the workflow only cares that the ordering is consistent. So it would require less brain overhead to understand how the estimation steps worked if we just made the feature vector a feature map instead, and dumped it to a vector or array (as needed) for use with JBlas.