如何避免Python 2.7类实例的深度复制？(How to avoid deepcopy of Python 2.7 class instances?)

 我正在运行一个演化算法来优化单独模拟环境中的函数（也用Python编写）。 模拟环境本身是一个python类实例，并且计算量很大（每次运行约4秒）并占用大约1Gb的内存。 由于我在这个模拟环境中生成大约十万个不同的函数来测试，我需要一种方法尽可能多地离线模拟。 
 
 我能做的是预先计算大部分模拟（即初始化类的一个实例，然后运行一些类函数的子集），然后在算法中仅针对一个小组件（例如，其余的几个函数）班上）。 这通常将评估时间缩短到约1秒，这显着改善了运行时间。 
 
 但是，每次将新函数传递到模拟类实例时，函数都会更改类实例（从而影响下一个函数求值的结果）。 我可以通过在每次评估之前对整个类实例进行深度复制来防止这种情况发生，但最终比每次运行整个模拟要慢。 
 
 有没有办法避免在每个阶段深入复制整个班级，但要获得相同的结果？ 基本上每个功能评估都需要在完全相同的类实例上执行。 
 
 谢谢 

I'm running an evolutionary algorithm to optimise functions within a separate simulation environment (also written in Python). The simulation environment itself is a python class instance, and is computationally expensive (~4 seconds per run) and takes up about 1Gb of memory. As I'm generating about a hundred thousand different functions to test in this simulation environment, I need a way to take as much of the simulation off-line as possible.
 
What I can do is pre-compute the majority of the simulation (i.e. initialise an instance of the class and then run some sub-set of class functions) and then fire in the algorithm for only a small component (e.g. the remaining few functions within the class). This in general reduces the evaluation time to ~1 sec, which improves the run-time significantly.
 
However, every time a new function is passed into the simulation class instance, the class instance is changed by the function (and thus affects the outcome of the next function evaluation). I can prevent this from happening by doing a deepcopy of the entire class instance before every evaluation, but that ends up being slower than running the entire simulation each time.
 
Is there a way to avoid deepcopying the entire class at each stage but to get the same result? Essentially each function evaluation needs to be performed on exactly the same class instance.
 
Thanks

原文：https://stackoverflow.com/questions/36796993