Haskell Backend
I have finally got the new backend written in Haskell into a useable
state. The sourcecode is available at
https://github.com/kirel/detexify-hs-backend . I am training the new
backend right now and will switch out the servers as soon as it is
finished. Internal benchmarks suggest a modest increase in recognition
accuracy and it should be faster too. I will keep the old server
around for a few days until I am sure that the new one performs fine.
to program in Haskell. I have also written a version in Clojure but
that one was too slow, probably due to my lack of understanding the
JVM. Haskell ist blazingly fast but of course I had a memory leak due
to excessive laziness.
I am still using nearest neighbor search but switched the distance
measure to a different algorithm. Strokes are now compared using a
greedy version of dynamic time warping.
