Not sure how scientific this comparison is, but I've got Max time used:
0.03/10.00, max memory used: 10264576/536870912.), and bytwigorlimerick
got Max time used: 0.04/10.00, max memory used: 10260480/536870912.).
https://courses.edx.org/courses/course-v1:UCSanDiegoX+ALGS200x+2T2017/discussion/forum/course/threads/5a4290cc44a15008b5000768
Looks like we are using different approaches.
Funny that my approach is a bit faster than

The only tricky part was to create and populate the 3-d list. The
instructions here was helpful:
https://www.ict.social/python/basics/multidimensional-lists-in-python

Not too bad, since I worked it out in Java already.

It turned out my Java test cases didn't cover enough number of cases. I
added one test case for it to fail, and figured out that the range in 2
loops should start with 1.

This isn't too bad after I worked out the algo in Java first.

I thought it wouldn't be this hard after I worked it out in Java first,
but once again it was fun and I learned a lot:
1. in all my previous exercises, my Python code performed better than my
Java code. This is the first time that my Java code significantly
outperforms my Python code: 0.92 seconds versus 7.93 seconds! My Java
code did use more than twice the memory.
2. Good re-enforcement of Python list copying concept, versus aliasing.
Don't forget the [:] (or [l:r]) magic!
3. list of (num, num) sort by first num by default, with proper tie
breaking. To sort on the second num, do list.sort(key=lambda x: x[1].
Cleve use of lambda, and it also does proper tie breaking.
4. I forgot to put this into my Java program comments, that Java code
gave me a good practice of comparator versus comparable interfaces,
which was very nice.

Fun stuff! Curious how much faster C++ program will be. I'll find out!

It wasn't hard since I worked this out in Java first. Once again, I felt
Python is really well designed and efficient.