编译器如何理解递归？(How do compilers understand recursion?)

 你的算法很好，但你的结果是不可代表的。 
 
 特别要注意的是，你的np.unique是无用的，因为每个load[i]包含重复项，所以结果的大小是列表长度乘以列表数目的乘积 
 
>>> np.prod([len(i) for i in second_example], dtype=np.int64) * 9
2498897217529908
 
 假设乐观地认为每个整数都是uint8 ，即2.2 PiB（Pebibytes），远远超过当前的RAM配置。 
 
 即使你不试图将整个结果一次写入内存，即使迭代这个过程也需要很长时间 - 假设一个慷慨的4GHz处理器和每个结果的单个时钟周期，你看起来比一个时间长一周完成 

Your algorithm is fine, but your result is unrepresentable.
 
Note in particular that your np.unique is useless, because each load[i] contains no duplicates, so the size of the result is the product of the lengths of the lists times the number of lists
 
>>> np.prod([len(i) for i in second_example], dtype=np.int64) * 9
2498897217529908
 
Assuming optimistically that each integer is a uint8, that's 2.2 PiB (Pebibytes), which far exceeds current RAM configurations.
 
Even if you don't try to put the whole result in memory at once, even iterating over this is going to take a long time - assuming a generous 4GHz processor and a single clock cycle per result, you're looking at longer than a week to finish