mieki256's diary

gem install narray-ruby19 --no-ri --no-rdoc

# NMatrix (NArray) のテスト

require 'benchmark'

require 'matrix' # 標準添付のMatrixライブラリ

require 'narray' # NArray

# require 'narray/narray' # SciRubyのNArray
# require 'nmatrix' # SciRubyのNMatrix

def get_proj_param(s, d)
  # x0, y0, x1, y1, x2, y2, x3, y3 = d.flatten.map {|item| item.to_f }
  # sx0, sy0, sx1, sy1, sx2, sy2, sx3, sy3 = s.flatten.map {|item| item.to_f }
  x0 = d[0][0].to_f
  x1 = d[1][0].to_f
  x2 = d[2][0].to_f
  x3 = d[3][0].to_f
  y0 = d[0][1].to_f
  y1 = d[1][1].to_f
  y2 = d[2][1].to_f
  y3 = d[3][1].to_f
  sx0 = s[0][0].to_f
  sx1 = s[1][0].to_f
  sx2 = s[2][0].to_f
  sx3 = s[3][0].to_f
  sy0 = s[0][1].to_f
  sy1 = s[1][1].to_f
  sy2 = s[2][1].to_f
  sy3 = s[3][1].to_f

  m = Matrix[
             [ sx0, sy0, 1.0, 0.0, 0.0, 0.0, -x0 * sx0, -x0 * sy0 ],
             [ sx1, sy1, 1.0, 0.0, 0.0, 0.0, -x1 * sx1, -x1 * sy1 ],
             [ sx2, sy2, 1.0, 0.0, 0.0, 0.0, -x2 * sx2, -x2 * sy2 ],
             [ sx3, sy3, 1.0, 0.0, 0.0, 0.0, -x3 * sx3, -x3 * sy3 ],
             [ 0.0, 0.0, 0.0, sx0, sy0, 1.0, -y0 * sx0, -y0 * sy0 ],
             [ 0.0, 0.0, 0.0, sx1, sy1, 1.0, -y1 * sx1, -y1 * sy1 ],
             [ 0.0, 0.0, 0.0, sx2, sy2, 1.0, -y2 * sx2, -y2 * sy2 ],
             [ 0.0, 0.0, 0.0, sx3, sy3, 1.0, -y3 * sx3, -y3 * sy3 ]
            ]

  n = m.inv # 逆行列を取得

  g_a = n[0,0] * x0 + n[0,1] * x1 + n[0,2] * x2 + n[0,3] * x3 + n[0,4] * y0 + n[0,5] * y1 + n[0,6] * y2 + n[0,7] * y3
  g_b = n[1,0] * x0 + n[1,1] * x1 + n[1,2] * x2 + n[1,3] * x3 + n[1,4] * y0 + n[1,5] * y1 + n[1,6] * y2 + n[1,7] * y3
  g_c = n[2,0] * x0 + n[2,1] * x1 + n[2,2] * x2 + n[2,3] * x3 + n[2,4] * y0 + n[2,5] * y1 + n[2,6] * y2 + n[2,7] * y3
  g_d = n[3,0] * x0 + n[3,1] * x1 + n[3,2] * x2 + n[3,3] * x3 + n[3,4] * y0 + n[3,5] * y1 + n[3,6] * y2 + n[3,7] * y3
  g_e = n[4,0] * x0 + n[4,1] * x1 + n[4,2] * x2 + n[4,3] * x3 + n[4,4] * y0 + n[4,5] * y1 + n[4,6] * y2 + n[4,7] * y3
  g_f = n[5,0] * x0 + n[5,1] * x1 + n[5,2] * x2 + n[5,3] * x3 + n[5,4] * y0 + n[5,5] * y1 + n[5,6] * y2 + n[5,7] * y3
  g_g = n[6,0] * x0 + n[6,1] * x1 + n[6,2] * x2 + n[6,3] * x3 + n[6,4] * y0 + n[6,5] * y1 + n[6,6] * y2 + n[6,7] * y3
  g_h = n[7,0] * x0 + n[7,1] * x1 + n[7,2] * x2 + n[7,3] * x3 + n[7,4] * y0 + n[7,5] * y1 + n[7,6] * y2 + n[7,7] * y3

  return [g_a, g_b, g_c, g_d, g_e, g_f, g_g, g_h]
end

def get_proj_param_nmatrix(s, d)
  x0 = d[0][0].to_f
  x1 = d[1][0].to_f
  x2 = d[2][0].to_f
  x3 = d[3][0].to_f
  y0 = d[0][1].to_f
  y1 = d[1][1].to_f
  y2 = d[2][1].to_f
  y3 = d[3][1].to_f
  sx0 = s[0][0].to_f
  sx1 = s[1][0].to_f
  sx2 = s[2][0].to_f
  sx3 = s[3][0].to_f
  sy0 = s[0][1].to_f
  sy1 = s[1][1].to_f
  sy2 = s[2][1].to_f
  sy3 = s[3][1].to_f

  # m = NArray::NMatrix[
  m = NMatrix[
              [ sx0, sy0, 1.0, 0.0, 0.0, 0.0, -x0 * sx0, -x0 * sy0 ],
              [ sx1, sy1, 1.0, 0.0, 0.0, 0.0, -x1 * sx1, -x1 * sy1 ],
              [ sx2, sy2, 1.0, 0.0, 0.0, 0.0, -x2 * sx2, -x2 * sy2 ],
              [ sx3, sy3, 1.0, 0.0, 0.0, 0.0, -x3 * sx3, -x3 * sy3 ],
              [ 0.0, 0.0, 0.0, sx0, sy0, 1.0, -y0 * sx0, -y0 * sy0 ],
              [ 0.0, 0.0, 0.0, sx1, sy1, 1.0, -y1 * sx1, -y1 * sy1 ],
              [ 0.0, 0.0, 0.0, sx2, sy2, 1.0, -y2 * sx2, -y2 * sy2 ],
              [ 0.0, 0.0, 0.0, sx3, sy3, 1.0, -y3 * sx3, -y3 * sy3 ]
                       ]


  n = m.inverse # 逆行列を取得

  g_a = n[0,0] * x0 + n[1,0] * x1 + n[2,0] * x2 + n[3,0] * x3 + n[4,0] * y0 + n[5,0] * y1 + n[6,0] * y2 + n[7,0] * y3
  g_b = n[0,1] * x0 + n[1,1] * x1 + n[2,1] * x2 + n[3,1] * x3 + n[4,1] * y0 + n[5,1] * y1 + n[6,1] * y2 + n[7,1] * y3
  g_c = n[0,2] * x0 + n[1,2] * x1 + n[2,2] * x2 + n[3,2] * x3 + n[4,2] * y0 + n[5,2] * y1 + n[6,2] * y2 + n[7,2] * y3
  g_d = n[0,3] * x0 + n[1,3] * x1 + n[2,3] * x2 + n[3,3] * x3 + n[4,3] * y0 + n[5,3] * y1 + n[6,3] * y2 + n[7,3] * y3
  g_e = n[0,4] * x0 + n[1,4] * x1 + n[2,4] * x2 + n[3,4] * x3 + n[4,4] * y0 + n[5,4] * y1 + n[6,4] * y2 + n[7,4] * y3
  g_f = n[0,5] * x0 + n[1,5] * x1 + n[2,5] * x2 + n[3,5] * x3 + n[4,5] * y0 + n[5,5] * y1 + n[6,5] * y2 + n[7,5] * y3
  g_g = n[0,6] * x0 + n[1,6] * x1 + n[2,6] * x2 + n[3,6] * x3 + n[4,6] * y0 + n[5,6] * y1 + n[6,6] * y2 + n[7,6] * y3
  g_h = n[0,7] * x0 + n[1,7] * x1 + n[2,7] * x2 + n[3,7] * x3 + n[4,7] * y0 + n[5,7] * y1 + n[6,7] * y2 + n[7,7] * y3

  return [g_a, g_b, g_c, g_d, g_e, g_f, g_g, g_h]
end

def dump_cof_diff(cof1, cof2)
  puts
  s = "abcdefgh"
  8.times do |i|
    d = cof1[i] - cof2[i]
    puts "#{s[i, 1]} = diff: #{d} , #{cof1[i]}, #{cof2[i]}"
  end
end


src = [[0, 0],[512, 0], [512, 512], [0, 512]]
dst = [ [160, 100], [320, 20], [600, 460], [100, 300]]

cof1 = []
cof2 = []
n = 10000
Benchmark.bm do |x|
  x.report("Array:  ") { n.times {|i| cof1 = get_proj_param(src, dst)} }
  x.report("NArray: ") { n.times {|i| cof2 = get_proj_param_nmatrix(src, dst) } }

  dump_cof_diff(cof1, cof2)
end

> ruby test_narray.rb
       user     system      total        real
Array:    3.042000   0.000000   3.042000 (  3.049174)
NArray:   0.421000   0.000000   0.421000 (  0.430025)

a = diff: -1.1102230246251565e-16 , 0.07848173515981738, 0.07848173515981749
b = diff: -5.551115123125783e-17 , -0.2090468036529681, -0.20904680365296804
c = diff: 0.0 , 160.0, 160.0
d = diff: 2.7755575615628914e-17 , -0.1708761415525114, -0.17087614155251143
e = diff: -5.551115123125783e-17 , 0.1150470890410959, 0.11504708904109595
f = diff: 0.0 , 100.0, 100.0
g = diff: 0.0 , -0.0007313070776255703, -0.0007313070776255703
h = diff: 0.0 , -0.0009185930365296805, -0.0009185930365296805

Ruby標準の Matrix を使うより、NArrayの NMatrix を使ったほうが、約7倍ほど速い、という結果になった。

ただ、Matrix と NMatrix は、添字の順番が逆になるらしい。そこは注意しないといかんようで。

_SciRuby/narray - GitHub を参考にインストール。

gem install narray-nmatrix --pre --no-ri --no-rdoc

NArray を使うか、SciRuby を使うかは、選べるらしい。

NArray を使う場合。

require 'narray'

SciRuby の NMatrix を使う場合。

require 'narray/narray'
require 'nmatrix'

名前空間？を使って指定する場合。

NArray::NMatrix   # NArray の NMatrix を使う場合
NMatrix           # SciRuby の NMatrix を使う場合