Just a notice that RFP for haystack, https://haystackconf.com/
, has gone
out. Proposals due jan 9th. Conference april 24th and 25th.
I'll probably submit the recent autocomplete work as a talk. Something like
the following. I don't like any of these titles yet and the abstract surely
needs help too, but will ponder on it.
Running millions of hits per second on a laptop
Elasticsearch and Tensorflow, an offline evaluation story
What if we could export the scoring equation along with all the inputs and
variables, and run this equation outside of elasticsearch? Simple queries
can then be run on a laptop at a rate of several million hits per second.
This opens up a variety of optimization techniques such as coordinate
ascent or hyperopt to tune the query variables including boosts, tie
breakers, similarity parameters, etc.