Dear team, I'm working on a project in which i

Question

Dear team, I'm working on a project in which i

have an imbalance dataset,
To train a model i did the train test split based on y, did the stratified cross validation for model on train dataset, did the hyperparameter tuning. Now i have finalized model with tuned hyperparameter

To build final model, i'll have to train finalzed model with tuned hyperparameter on full data set which is original train + test.

Now my question is how should i give data to the model.
Do i have to provide original dataset which is imbalanced? or do i have to give balanced original dataset?

#english #programming

0

07.08.2021

3 ответов

40 просмотров

Pablo

You give the data you used for hyperparameter tuning. Otherwise, you are giving another data, as this will be modified (balaned)

0

09.08.2021

Lokesh Автор вопроса

Yes, thank you. I have been thinking the same thing that i should use same data. Otherwise what is the point of doing hyperparameter tuning and cross validation. Any idea how can i see how much final model performance is deviated from cross validation score?

0

10.08.2021

Maria Lavrovskaya · Accepted Answer

Maria Lavrovskaya

In my opinion, there is no yes/no kind of answer. If you’ve done stratified train/test split based on the feature you are trying to predict, then the final model where train+test is used as train would be imbalanced by setup. If you’ve employed over/under sampling techniques up front, then this is where you’ve done some balancing and thus I doubt the model would generalize that well

0

08.08.2021

169 похожих чатов

Dear team, I'm working on a project in which i

3 ответов

Похожие вопросы