Python / numpy > QMC(Halton_sequence)に基づく要素の取り出しの実装

動作環境

Ubuntu 14.04 LTS desktop amd64
GeForce GTX 750 Ti
ASRock Z170M Pro4S [Intel Z170chipset]
TensorFlow v0.11
cuDNN v5.1 for Linux
CUDA v7.5
Python 2.7.6
IPython 5.1.0 -- An enhanced Interactive Python.

TensorFlowにおいてmnist.next_batch()の代わりに使う予定のQMCコード。
小さなコードで動作確認をしてみた。

imglistからHalton Sequenceを用いて要素を取り出す。
関連 http://qiita.com/7of9/items/7a8920531fd715d0b75a
関連 http://qiita.com/7of9/items/9db1251e714952794023

v0.1

sample.py

import numpy as np

imglist = np.array([10,20,30,40,50,60,70,80,90])

def Halton_sequence(i0):
    xbase = 2
    ybase = 3

    invxbase = 1.0 / xbase
    facx = 1.0 / xbase

    invybase = 1.0 / ybase
    facy = 1.0 / ybase

    inp = i0
    x0 = 0.0
    while inp > 0:
        x0 = x0 + (inp % xbase) * invxbase
        inp = inp / xbase
        invxbase = invxbase * facx

    inp = i0
    y0 = 0.0
    while inp > 0:
        y0 = y0 + (inp % ybase) * invybase
        inp = inp / ybase
        invybase = invybase * facy

    return x0, y0

length = len(imglist)

idx = np.array([])
for i0 in range(10):
	# discard 2nd
	res, _ = Halton_sequence(i0)
	# print res * length
	idx = np.append(idx, res * length)

print idx
print imglist[idx.astype(int)]

結果

$ python sample.py 
[ 0.      4.5     2.25    6.75    1.125   5.625   3.375   7.875   0.5625
  5.0625]
[10 50 30 70 20 60 40 80 10 60]

v0.2 QMC_getIndex()追加

sample.py

'''
v0.2 2016 Oct. 22
	- add QMC_getIndex()
v0.1 2016 Oct. 22
	- use Halton_sequence() together with unorganized code
'''

import numpy as np

imglist = np.array([10,20,30,40,50,60,70,80,90])

def Halton_sequence(i0):
    xbase = 2
    ybase = 3

    invxbase = 1.0 / xbase
    facx = 1.0 / xbase

    invybase = 1.0 / ybase
    facy = 1.0 / ybase

    inp = i0
    x0 = 0.0
    while inp > 0:
        x0 = x0 + (inp % xbase) * invxbase
        inp = inp / xbase
        invxbase = invxbase * facx

    inp = i0
    y0 = 0.0
    while inp > 0:
        y0 = y0 + (inp % ybase) * invybase
        inp = inp / ybase
        invybase = invybase * facy

    return x0, y0

def QMC_getIndex(listSize, start, getsize):
	idx = np.array([])
	for i0 in range(start, start + getsize):
		res, _ = Halton_sequence(i0) # discard 2nd dim
		idx = np.append(idx, res * listSize)
	return idx

length = len(imglist)
idx = QMC_getIndex(length, 0, 10)
print "idx:",idx
print "img:",imglist[idx.astype(int)]

idx = QMC_getIndex(length, 10, 10)
print "idx:",idx
print "img:",imglist[idx.astype(int)]

結果

$ python sample.py 
idx: [ 0.      4.5     2.25    6.75    1.125   5.625   3.375   7.875   0.5625
  5.0625]
img: [10 50 30 70 20 60 40 80 10 60]
idx: [ 2.8125   7.3125   1.6875   6.1875   3.9375   8.4375   0.28125  4.78125
  2.53125  7.03125]
img: [30 80 20 70 40 90 10 50 30 80]