Question

我正在尝试优化以下MATLAB代码的时间效率，它目前需要超过4个小时才能运行（我已经预先分配了两个结构，但这里没有包含该部分）：

for combination = 1:1771

  for hankel_size = 1:4;

    for window = 1:999

      Output.bin_r(:, window, combination, hankel_size) = bsxfun(@minus, data.hankel_index_mean(window, combination ,hankel_size),centers(window, :, hankel)');
      Output.score(window, combination, hankel_size) = probs(window, :, hankel_size)*Output.bin_r(:, window, combination, hankel_size);

    end

  end

end

请注意：

center是一个999 x 50 x 4矩阵
hankel_index_mean是一个999 x 1771 x 4矩阵
probs是一个999 x 50 x 4矩阵

提前感谢您的帮助！

Answer 1

parfor combination = 1:1771

  for hankel_size = 1:4;

    for window = 1:999

      Output.bin_r(:, window, combination, hankel_size) = bsxfun(@minus, data.hankel_index_mean(window, combination ,hankel_size),centers(window, :, hankel)');
      Output.score(window, combination, hankel_size) = probs(window, :, hankel_size)*Output.bin_r(:, window, combination, hankel_size);

    end

  end

end

parfor利用CPU中的所有核心。

通过matlab默认打开并行池，默认在调用并行函数（例如parfor或spmd）时打开一个并通过调用parpool或{{1 }}

在Home-＆gt; parallel-＆gt; parallel preferences下编辑并行首选项。

试图将以下内容用于时间效率MATLAB

1 个答案: