Fixing flaky test

[crawlingcub]

I ran your test-suite by removing all seed settings (such as np.random_seed(0)) and found that some tests fail quite often. For instance, the test test_select_best_model_ranking_random failed 3 out of 500 times. Each time the same assertion failed.

To fix the test, I modified the assertion to ensure that it does not fail if the seeds are removed (or if the seed/rng implementation is changed in the future). The test does not fail now at all without seeds.

Please let me know if this fix seems reasonable. If yes, I can also look into other tests which failed and try to make them less flaky. I will be happy to incorporate any other changes that you guys may suggest.

Thanks!