27 Nov
2009
27 Nov
'09
9:11 p.m.
Am Samstag 28 November 2009 02:04:31 schrieb Daniel Fischer:
Make it
distance orig new = f m n
and no thunks need be written at all in this case. Cuts down running time by nearly half :)
I think you could speed it up significantly by calculating the distance more lazily.
Yup :D Timings here: Your last ByteString code: 128s That with corrected index calculation: 172s Correct indices and distance orig new = f m n instead of memf m n: 99s That and an array for the substitution cost: 64s String IO, substitution cost array and a suitably lazy break-early distance function: 1.7s That and a lazier surcharge function: 1.5s The same with ByteString IO: 1.4s Yes, I'd say that qualifies as a significant speedup.