Speedup our eight slowest pytests (one at a time please)

[cclauss]

Feature description

At the end of our GitHub Actions build jobs, there is a list of the slowest ones.

Are there ways to speed up these tests without reducing our functionality or our code coverage?

Please only fix one algorithm per pull request.

• 16.64s call neural_network/simple_neural_network.py::neural_network.simple_neural_network.forward_propagation
• 6.15s call backtracking/word_search.py::backtracking.word_search.word_exists
• 4.61s call digital_image_processing/test_digital_image_processing.py::test_local_binary_pattern
• 2.59s call backtracking/power_sum.py::backtracking.power_sum.backtrack
• 2.19s call machine_learning/xgboost_regressor.py::machine_learning.xgboost_regressor.main
• 1.73s call maths/prime_numbers.py::maths.prime_numbers.slow_primes
• 1.43s call backtracking/power_sum.py::backtracking.power_sum.solve
• 1.00s call other/dijkstra_bankers_algorithm.py::other.dijkstra_bankers_algorithm.BankersAlgorithm.main
  ================= 1506 passed, 25 warnings in 71.23s (0:01:11) =================

Also, those 25 pytest warnings are worth fixing!!!

[tianyizheng02]

I pointed out in #8785 that neural_network/simple_neural_network.py has multiple issues (I don't think it even implements a neural network) and could do with a rewrite. Ideally whatever new implementation replaces it should run a lot more quickly.

[Muhammadummerr]

I checked the maths/prime_numbers.py file. In this file, there are three methods for finding prime numbers using different techniques. The simplest method, referred to as slow_prime (which uses a loop to iterate from 2 to n and print prime numbers), is taking a lot of time. Do we need it there? Since the other two methods are faster with just minor modifications for finding prime numbers, do we still need the slow_prime method in there?

[cclauss]

Can slow_prime() be tested with smaller numbers? We do not want to get rid of it but we want its test to finish faster.

Python/maths/prime_numbers.py

Lines 20 to 21 in 87494f1

	>>> list(slow_primes(10000))[-1]
	9973

Let's lower the 10_000 to 1_000

[Muhammadummerr]

If 'slow_prime()' is tested on small test cases, then the other two methods will also be tested on small test cases to compare the time taken among them.

[Muhammadummerr]

If I reduce the test cases, do I need to raise a PR to check whether they finished faster, or is there any other method I can use to verify if the issue is resolved before submitting a PR?

[cclauss]

When things are run on CI platforms like GitHub Actions then an environment variable CI is defined.

• https://docs.github.com/en/actions/learn-github-actions/variables#default-environment-variables

Can you use https://docs.python.org/3/library/os.html#os.getenv to see if CI is defined? If CI is defined then set the number to 1_000 for all three tests and it if is not defined then set it to 10_000 for all three.

[Muhammadummerr]

Okay, thanks. Let me try.

[Muhammadummerr]

But if it is not defined, setting the test case to 10_000 will take the same amount of time as before, doesn't it?

[cclauss]

Yes, but we do not care because it no longer slows down our build process.

[Muhammadummerr]

I sped it up in #9851.

[Muhammadummerr]

I was reviewing the backtracking/power_sum.py code, and I noticed that the main issue causing slow execution is the large value of 'needed_sum' parameter.To improve performance, is it correct to reduce the 'needed_sum' value?

[cclauss]

@duongoku Would you be willing to advise @Muhammadummerr on the best way to speed up the slow doctests in backtracking/power_sum.py ?

[Muhammadummerr]

I would love to hear from @duongoku.