Fix random FEDOT results with set seed #1143
Conversation
|
Hello @kasyanovse! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found: There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2023-08-22 13:17:20 UTC |
Codecov Report
@@ Coverage Diff @@
## master #1143 +/- ##
==========================================
- Coverage 78.48% 78.32% -0.17%
==========================================
Files 130 130
Lines 9323 9350 +27
==========================================
+ Hits 7317 7323 +6
- Misses 2006 2027 +21
|
Model become less random but it is not full solution
kasyanovse
force-pushed
the
bugfix-1128
branch
from
7c9011d
to
43ec51d
Compare
| assert np.allclose(fedot.history.all_historical_fitness, old_fedot.history.all_historical_fitness) | ||
| assert np.isclose(np.sum(np.abs(fedot.forecast(test) - old_fedot.forecast(test))), 0) | ||
|
|
||
| old_fedot = fedot |
There was a problem hiding this comment.
А тут не нужно deecopy? Мы же потом меняем fedot.
There was a problem hiding this comment.
Мне кажется нет. Когда вызываем конструктор федота, то он в fedot просто ссылку на новый инстанс запишет, а ссылка на старый останется в old_fedot.
There was a problem hiding this comment.
Может добавить тогда ветку теста, где seed меняется и проверка идет на то результаты не идентичны?
Для уверенности.
test/integration/real_applications/test_model_result_reproducing.py
Outdated
Show resolved
Hide resolved
FEDOT with set seed return the same pipeline for different compose processes for now.