2.5x faster

[vprelovac]

Performance optimization, about 2.5x faster

[miso-belica]

Thanks for the changes. Good job 👍

I added some comments so please try to react on them (by code or another comment :) and if it would be possible could you maybe create some test for method _create_matrix so we are sure it's the same as before?

Oh, and I almost forgot. If you have any measurements/benchmark to show how the changes result in the 2.5x speedup it would be great if you can provide that too.

[miso-belica]

I believe you should do range(0, sentences_count) here because list(range(0, 3)) == [0, 1, 2]. But I would prefer you return enumerate(sentences_as_words). I think the one last missing word may a reason why the test fails on CI. So the code after the changes will be:

for i, words_i in enumerate(sentences_as_words):
  for j in range(i+1, sentences_count):
    rating = self._rate_sentences_edge(words_i, sentences_as_words[j])
    weights[i, j] = rating
    weights[j, i] = rating

EDIT: now I realized, this is not the case when you reach the end you fill the last row by weights[j, i] = rating. But still, the test fails so please check it why.

[miso-belica]

Don't have time to do these, feel free to reject and implement changes yourself.

@vprelovac Thanks for the inspiration. I implemented the optimizations in the related PR but was not so aggressive so the tests pass and it is working the same as before.

But still, the problem was in the diagonal of the matrix which should not affect the score so I will dig deeper into it to find out why we can't omit the computing of the diagonal.

Have a nice day :)