Rewrote get_top_k_items() to improve runtime

[ChuyangKe]

Description

Rewrote get_top_k_items().

Optimize the runtime of ranking metrics.

Related Issues

Checklist:

• I have followed the contribution guidelines and code style for this project.
• I have added tests covering my contributions.
• I have updated the documentation accordingly.
• This PR is being made to staging branch and not to main branch.

[gramhagen]

Do you have any timing results on this change?

[ChuyangKe]

Do you have any timing results on this change?

I tested the old vs new function on the movielens dataset with 1m samples. I set k = 10.

Here are the results:
original_time: 5.363674836989958 seconds.
new_time: 0.12664664999465458 seconds.

[simonzhaoms]

@ChuyangKe Could you please also add a test case for get_top_k_items() in tests/unit/recommenders/evaluation/test_python_evaluation.py? I find there is no test for get_top_k_items(). I think you can use the data from rating_true for the test.

[miguelgfierro]

This is good progress @ChuyangKe. I agree with Simon, it would be great to have the tests. This can be done in this PR or in a different one, depending on what you prefer. Please let Simon and me know.

In another PR you maybe can explore how to perform the same operation with numpy, hopefully this gives us a much better time.
Some code I found in the web:

• with argpartition: https://stackoverflow.com/questions/43386432/how-to-get-indexes-of-k-maximum-values-from-a-numpy-multidimensional-array
• groupby with numpy (maybe the other method is better): https://cmdlinetips.com/2019/05/how-to-implement-pandas-groupby-operation-with-numpy/

[ChuyangKe]

This is good progress @ChuyangKe. I agree with Simon, it would be great to have the tests. This can be done in this PR or in a different one, depending on what you prefer. Please let Simon and me know.

In another PR you maybe can explore how to perform the same operation with numpy, hopefully this gives us a much better time. Some code I found in the web:

• with argpartition: https://stackoverflow.com/questions/43386432/how-to-get-indexes-of-k-maximum-values-from-a-numpy-multidimensional-array
• groupby with numpy (maybe the other method is better): https://cmdlinetips.com/2019/05/how-to-implement-pandas-groupby-operation-with-numpy/

I will create another PR for the test case @simonzhaoms @miguelgfierro.

Just wanted to make sure @miguelgfierro, do you mean implementing the same function get_top_k_items() using numpy (instead of current pandas operations), or is it another function?

[miguelgfierro]

do you mean implementing the same function get_top_k_items() using numpy (instead of current pandas operations), or is it another function?

get_top_k_items() using numpy

I'll merge this, super good @ChuyangKe! keep up the good work