UNPKG

dataship-frame

Version:

A Data Frame for Javascript. Crunch numbers in node and the browser.

github.com/dataship/frame

dataship/frame

36 lines (22 loc) • 2.01 kB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
### Why are my dv results not consistent with their benchmark webpage? because it slows down with consecutive runs, dropping to a quarter of initial performance. ok 1 table.query.sum: 1000000x3 # 12.019 MFlops/sec ±16.51% n = 15 µ = 83ms : [0.022,0.02225,0.0935,0.092,0.0925,0.0925,0.09325,0.092,0.093,0.09275,0.09275,0.09225,0.09275,0.092,0.0925] ### Can I make Frame as fast as dv by encoding the strings? likely it will give a 3x speedup. #### integers ok 1 groupby.sum: 1000000x3 # 13.952 MFlops/sec ±2.01% n = 29 µ = 72ms : [0.0645,0.0625,0.0635,0.062,0.0775,0.0725,0.0715,0.073,0.0725,0.073,0.0735,0.0745,0.0725,0.0725,0.074,0.073,0.0715,0.077,0.072,0.071,0.072,0.0715,0.0725,0.0735,0.0725,0.073,0.0745,0.0715,0.0735] #### strings ok 1 groupby.sum: 1000000x3 # 4.120 MFlops/sec ±3.74% n = 14 µ = 243ms : [0.239,0.235,0.232,0.234,0.267,0.267,0.24,0.235,0.235,0.236,0.279,0.233,0.233,0.233] ### Is the FrameIndex.reduce faster than dv.query, when Frame.groupby has already been run? yes, but not quite faster than the ultra-fast first two runs of dv.query ok 1 sum: 1000000x3 # 23.298 MFlops/sec ±1.40% n = 34 µ = 43ms : [0.037333333333333336,0.042333333333333334,0.042333333333333334,0.042666666666666665,0.042333333333333334,0.042666666666666665,0.042333333333333334,0.043333333333333335,0.042666666666666665,0.04566666666666667,0.044000000000000004,0.042666666666666665,0.044333333333333336,0.042666666666666665,0.04733333333333333,0.043666666666666666,0.042,0.042333333333333334,0.043000000000000003,0.042333333333333334,0.042666666666666665,0.042,0.044000000000000004,0.042666666666666665,0.042666666666666665,0.042666666666666665,0.042,0.042333333333333334,0.043000000000000003,0.048666666666666664,0.041666666666666664,0.041666666666666664,0.042333333333333334,0.043000000000000003] ### Can I make FrameIndex.reduce faster than the ultra-fast dv.query? try: 1. reproducing results 2. removing the function call ### Is the dv setup longer? ### Why is dv faster initially?