python で numpy の値を swap する方法
前提
get_idxs関数は、ランダムな２つのインデックスを返します
結論が知りたい方はまとめまで飛ばしてください
方法１
C言語的な記述法（たぶん誰もが最初に思いつく）
def swap1(arr):
idx0, idx1 = get_idxs(arr)
tmp = arr[idx0]
arr[idx0] = arr[idx1]
arr[idx1] = tmp
方法２
説明しずらい
def swap2(arr):
idx0, idx1 = get_idxs(arr)
arr[[idx0, idx1]] = arr[[idx1, idx0]]
方法３
タプルで
def swap3(arr):
idx0, idx1 = get_idxs(arr)
arr[idx0], arr[idx1] = arr[idx1], arr[idx0]
ベンチマークテスト
せっかくなので計測っちゃいましょうと
ズラリ
## benchmarker: release 4.0.1 (for python)
## python version: 3.7.1
## python compiler: GCC 7.3.0
## python platform: Linux-4.15.0-45-generic-x86_64-with-debian-buster-sid
## python executable: /opt/anaconda3/bin/python
## cpu model: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz # 3504.115 MHz
## parameters: loop=1000000, cycle=10, extra=1
## (#1) real (total = user + sys)
(Empty) 0.0156 0.0100 0.0100 0.0000
swap1 2.3436 2.3500 2.3500 0.0000
swap2 3.7167 3.7200 3.7200 0.0000
swap3 2.3666 2.3800 2.3800 0.0000
## (#2) real (total = user + sys)
(Empty) 0.0154 0.0100 0.0100 0.0000
swap1 2.3326 2.3400 2.3400 0.0000
swap2 3.6803 3.6800 3.6800 0.0000
swap3 2.3584 2.3700 2.3700 0.0000
## (#3) real (total = user + sys)
(Empty) 0.0157 0.0100 0.0100 0.0000
swap1 2.3131 2.3200 2.3200 0.0000
swap2 3.6810 3.6900 3.6900 0.0000
swap3 2.3164 2.3200 2.3200 0.0000
## (#4) real (total = user + sys)
(Empty) 0.0177 0.0200 0.0200 0.0000
swap1 2.3225 2.3200 2.3200 0.0000
swap2 3.6397 3.6300 3.6300 0.0000
swap3 2.3224 2.3200 2.3200 0.0000
## (#5) real (total = user + sys)
(Empty) 0.0152 0.0200 0.0200 0.0000
swap1 2.3206 2.3100 2.3100 0.0000
swap2 3.6659 3.6600 3.6600 0.0000
swap3 2.3865 2.3900 2.3900 0.0000
## (#6) real (total = user + sys)
(Empty) 0.0165 0.0100 0.0100 0.0000
swap1 2.3145 2.3200 2.3200 0.0000
swap2 3.6422 3.6500 3.6500 0.0000
swap3 2.3354 2.3400 2.3400 0.0000
## (#7) real (total = user + sys)
(Empty) 0.0162 0.0200 0.0200 0.0000
swap1 2.2964 2.2900 2.2900 0.0000
swap2 3.6841 3.6800 3.6800 0.0000
swap3 2.3225 2.3200 2.3200 0.0000
## (#8) real (total = user + sys)
(Empty) 0.0151 0.0100 0.0100 0.0000
swap1 2.3031 2.3100 2.3100 0.0000
swap2 3.6650 3.6700 3.6700 0.0000
swap3 2.3700 2.3800 2.3800 0.0000
## (#9) real (total = user + sys)
(Empty) 0.0158 0.0100 0.0100 0.0000
swap1 2.3015 2.3100 2.3100 0.0000
swap2 3.6395 3.6400 3.6400 0.0000
swap3 2.3125 2.3200 2.3200 0.0000
## (#10) real (total = user + sys)
(Empty) 0.0153 0.0200 0.0200 0.0000
swap1 2.3200 2.3100 2.3100 0.0000
swap2 3.6493 3.6500 3.6500 0.0000
swap3 2.3936 2.3800 2.3800 0.0000
## (#11) real (total = user + sys)
(Empty) 0.0149 0.0200 0.0200 0.0000
swap1 2.3522 2.3500 2.3500 0.0000
swap2 3.7416 3.7300 3.7300 0.0000
swap3 2.3533 2.3500 2.3500 0.0000
## (#12) real (total = user + sys)
(Empty) 0.0159 0.0200 0.0200 0.0000
swap1 2.3102 2.3000 2.3000 0.0000
swap2 3.6899 3.6900 3.6900 0.0000
swap3 2.3286 2.3200 2.3200 0.0000
## Ignore min & max min cycle max cycle
swap1 2.2964 (#7) 2.3522 (#11)
swap2 3.6395 (#9) 3.7416 (#11)
swap3 2.3125 (#9) 2.3936 (#10)
## Average of 10 (=12-2*1) real (total = user + sys)
swap1 2.3182 2.3190 2.3190 0.0000
swap2 3.6714 3.6720 3.6720 0.0000
swap3 2.3460 2.3490 2.3490 0.0000
## Ranking real
swap1 2.3182 (100.0) ********************
swap3 2.3460 ( 98.8) ********************
swap2 3.6714 ( 63.1) *************
## Matrix real [01] [02] [03]
[01] swap1 2.3182 100.0 101.2 158.4
[02] swap3 2.3460 98.8 100.0 156.5
[03] swap2 3.6714 63.1 63.9 100.0
意外や意外！
なんと 方法１が一番はやい 結果になりました
次が方法３、最後に方法２と続きます
方法１と方法３がいい勝負してますね
方法２は安定して遅いようです
まとめ
文法が簡単かつ実行速度のはやい方法３を使いましょう
ソースコード
bm.py
from numpy import arange
from benchmarker import Benchmarker
import swap
LOOP = 1000000
ARR_LEN = 2000
CYC = 10
with Benchmarker(LOOP, cycle=CYC, extra=1) as bench:
n = arange(ARR_LEN)
@bench(None) ## empty loop
def _(bm):
for i in bm:
pass
@bench("swap1")
def _(bm):
for i in bm:
swap.swap1(n)
@bench("swap2")
def _(bm):
for i in bm:
swap.swap2(n)
@bench("swap3")
def _(bm):
for i in bm:
swap.swap3(n)
swap.py
from numpy.random import randint
def swap1(arr):
idx0, idx1 = get_idxs(arr)
tmp = arr[idx0]
arr[idx0] = arr[idx1]
arr[idx1] = tmp
def swap2(arr):
idx0, idx1 = get_idxs(arr)
arr[[idx0, idx1]] = arr[[idx1, idx0]]
def swap3(arr):
idx0, idx1 = get_idxs(arr)
arr[idx0], arr[idx1] = arr[idx1], arr[idx0]
def get_idxs(arr):
return randint(arr.shape[0]), randint(arr.shape[0])
bm.py
from numpy import arange
from benchmarker import Benchmarker
import swap
LOOP = 1000000
ARR_LEN = 2000
CYC = 10
with Benchmarker(LOOP, cycle=CYC, extra=1) as bench:
n = arange(ARR_LEN)
@bench(None) ## empty loop
def _(bm):
for i in bm:
pass
@bench("swap1")
def _(bm):
for i in bm:
swap.swap1(n)
@bench("swap2")
def _(bm):
for i in bm:
swap.swap2(n)
@bench("swap3")
def _(bm):
for i in bm:
swap.swap3(n)
swap.py
from numpy.random import randint
def swap1(arr):
idx0, idx1 = get_idxs(arr)
tmp = arr[idx0]
arr[idx0] = arr[idx1]
arr[idx1] = tmp
def swap2(arr):
idx0, idx1 = get_idxs(arr)
arr[[idx0, idx1]] = arr[[idx1, idx0]]
def swap3(arr):
idx0, idx1 = get_idxs(arr)
arr[idx0], arr[idx1] = arr[idx1], arr[idx0]
def get_idxs(arr):
return randint(arr.shape[0]), randint(arr.shape[0])
Author And Source
この問題について(python で numpy の値を swap する方法), 我々は、より多くの情報をここで見つけました https://qiita.com/chankane/items/80244c32f18eff609fa4著者帰属：元の著者の情報は、元のURLに含まれています。著作権は原作者に属する。
Content is automatically searched and collected through network algorithms . If there is a violation . Please contact us . We will adjust (correct author information ,or delete content ) as soon as possible .