在 Mathematica 中并行评估单元
我有数百个单元格长的 Mathematica 文件，我想使用并行计算。我的每台机器都有 2 个处理器 x 4 核，内存为 16 Gb。我的 Mathematica 许可证允许我运行最多 2 个主内核，其中 1 个主内核可以有 4 个从内核（这是我玩了一段时间后的解释）。
我曾经在两个不同笔记本的两个主内核中运行我的代码。为了进一步加快速度，我尝试使用 ParallelEvaluate[] 封装一些单元格，它似乎有效。然后，我还有 4 个代码副本通过主内核之一运行，彼此不知情，这很好。 （我基本上是在尝试并行运行尽可能多的代码/数学内核副本。我还没有追求任何真正并行的东西）。
由于我的代码太长且复杂，我不想再次编辑每个单元格以使它们并行计算。有没有什么神奇的东西可以放在我的笔记本的开头，这样之后评估的每个单元格默认都是ParallelEvaluate[...单元格内容....]？
I have this many hundreds of cell long Mathematica file and I want to use parallel evaluation. I have a 2 processor x 4 core each machine with 16 Gb memory. My Mathematica license allows me to run at most 2 master kernels with 1 of the master could have 4 slave kernels (this is my interpretation after I played with it for a while).
I used to run my code in two master kernels in two different notebooks. To speed up things further, I tried to encapsulate a few cells with ParallelEvaluate[] and it seemed to work. Then I also have 4 copies of my code running unaware of each other through one of the master kernels, which is fine. (I am basically trying to run as many copies of my code/mathkernel in parallel as possible. I am not shooting for anything truly parallel yet).
Since my code is too long and complicated, I do not want to edit every cell again to make them evaluate in parallel. Is there anything magical I can put in the beginning of my notebook so every cell evaluated after that will be by default ParallelEvaluate[ ... cell contents.... ]?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
正如 belisarius 所建议的， $Pre=ParallelEvaluate 正是我想要的。一个问题是，当我执行 $Pre=Identity 将其关闭以便我可以返回主内核时，mathematica 仍然尝试在从内核而不是主内核中对其进行评估，并且失败。我最终解决如下： SetSharedVariable[parallelcontrol];并行控制=并行评估； $Pre := 并行控制; ...这里的一切都是在奴隶中评估的...;并行控制=同一性； ....一切都返回到仅在主机上进行评估...这是在我的笔记本电脑上运行的示例，它有 2 个内核：
{KernelObject[1, "local"], KernelObject[2, "local"]}
0
{ 1, 2}
Null[ParallelEvaluate]
{1, 2}
{身份, 身份}
0
ParallelEvaluate
{1, 2}
As suggested by belisarius, $Pre=ParallelEvaluate does exactly what I want. One problem was when I do $Pre=Identity to turn it off so I can go back to my master kernel, mathematica still tries to evaluate that in slave kernels instead of master and fails. I ended up solving it as follows: SetSharedVariable[parallelcontrol]; parallelcontrol = ParallelEvaluate; $Pre := parallelcontrol; ... everything is evaluated in slaves here ... ; parallelcontrol = Identity; .... everything go back to be evaluated on master only ... Here is a sample run on my laptop which has 2 cores:
{KernelObject[1, "local"], KernelObject[2, "local"]}
0
{1, 2}
Null[ParallelEvaluate]
{1, 2}
{Identity, Identity}
0
ParallelEvaluate
{1, 2}