The primary goal of this pipeline is to take an elasticsearch query, run it with explain enabled, and collect the hits for some number of query strings. It should be able to extract all the scoring components from the explain into a single dict per hit, and an explain must be able to convert
into a tensorflow graph that will, when given the input feature vectors, score the hits.
At a high level the idea is to be able to run the scoring equation outside elasticsearch with all the components of the equation pre-calculated. Variables in the equations can then be tuned and the results re-run without having to query elasticsearch directly. In testing some simple queries can run at 1M hits/sec on modest hardware which opens up more tuning possibilities.