3
$\begingroup$

I did a stacking using three base classifiers RF, NB, KN N and metamodel random forest or SVM using sklearn library

But which is strange each time i change the metamodel i got the same results. Is it normal ?????

$\endgroup$
  • $\begingroup$ @SeanOwen may you advise to free the question it is an intersiting question and i gave all details $\endgroup$ – Rawia Sammout Nov 21 '18 at 11:39
  • $\begingroup$ I don't mind reopening it; it's borderline. Really it's better to boil this down to the essential piece of code you're asking about, and give more detail about what you have observed, narrowing it down further. $\endgroup$ – Sean Owen Nov 21 '18 at 19:16
  • $\begingroup$ Thanks for your comment i found it an interesting question about a stacking principle and it is not about code So i put all the details and explanation as required with an example with iris dataset no further details to add thanks for your kind support i see people here evaluate the effort of others without even asking i asked to demand explanation $\endgroup$ – Rawia Sammout Nov 22 '18 at 6:55
3
$\begingroup$

No, in generally speaking, even minor changes should affect your performance. Changing your meta-model should normally have a visible impact in your model's performance.

Two things you can try:

  • Check for any problems in your code.
  • Maybe your test set size is really small. For example if you have 5 test samples, it isn't difficult for all models to get 4/5 (i.e. 80% accuracy). As your test size increases so should the variance of your models' performance.
| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.