None
EN
SimdSketch: a fast bucket sketch
['Ragnar', 'Groot Koerkamp']
home on CuriousCoding
Instead of pulling values from filled buckets to empty bucket, filled buckets can push values into empty buckets (Mai et al. 2020). In that case, it’s faster to push from filled buckets to empty buckets: each bucket tries copying its value to empty buckets util all are filled. Tool Sketch type size \(s\) $b$-bits Sketching (s) Distances (s) Correlation BinDash bottom 128 64 113.3 0.580 0.9669 BinDash bottom 1024 64 116.3 3.580 0.9960 BinDash bottom 8192 64 138.6 29.450 0.9994 SimdSketch bottom 128 8 2.0 0.216 0.9655 SimdSketch bottom 1024 8 2.0 1.699 0.9965 SimdSketch bottom 8192 8 2.1 13.689 0.9993 SimdSketch bottom 32768 8 2.8 54.954 0.9999 BinDash bucket 128 1 124.7 0.270 0.8996 BinDash bucket 128 8 124.7 0.250 0.9649 BinDash bucket 128 32 124.6 0.240 0.9650 BinDash bucket 1024 1 125.2…