@carette the 8 second gap the @rjnw refers to is between the printouts in the Haskell run

Which includes considering 100 array elements and updating about 10 of them

So it takes 8s to update 10 array elements? That seems, er, excessive?

8s is for calling update 10 times, which is calling the prog function which is generated by hakaru. (Sorry for confusing terminology, update is the name of curried prog in the code written for benchmarking)

which means around 0.8s in haskell compared to 0.26s with sham for calling prog once.

I don’t actually know how the code in hk runner does, the runner calling update 10 times was what we found out in today’s meeting but I was running it again with my own code wrapper around the prog function to compare haskell and sham outputs. For me haskell takes 8.5seconds just for one update. https://github.com/hakaru-dev/hakaru-rktjit/blob/master/test/NaiveBayes/hs/MainProg.hs here is the code I used. I am not sure if I am doing anything wrong or if the runner code for NaiveBayes is calling update only once between 100 updates.