Do you need (or can you) use split or cross- validation if you're using Deep Learning?

Curious
Curious New Altair Community Member
edited November 2024 in Community Q&A
Would it be correct to use deep learning in training partition?

Best Answers

  • varunm1
    varunm1 New Altair Community Member
    Answer ✓
    Hi @Curious

    You can use cross-validation for deep learning as the results are consistent compared to split validation. Personally, I use cross-validation for every model I develop as this gives confidence about the results.

    Thanks
  • SGolbert
    SGolbert New Altair Community Member
    Answer ✓

    it is correct, you can take a look at Auto Model with model type "Deep Learning" to have an example. It is always better to do cross validation, and maybe reduce the folds if it takes too long.

    It is good to remark that cross validation is used only for measuring the model performance with a reasonable bias. Depending of whether you need to optimize hyperparameters or not, cross validation may not be needed in a production environment.

    Regards,
    Sebastian

Answers

  • varunm1
    varunm1 New Altair Community Member
    Answer ✓
    Hi @Curious

    You can use cross-validation for deep learning as the results are consistent compared to split validation. Personally, I use cross-validation for every model I develop as this gives confidence about the results.

    Thanks
  • SGolbert
    SGolbert New Altair Community Member
    Answer ✓

    it is correct, you can take a look at Auto Model with model type "Deep Learning" to have an example. It is always better to do cross validation, and maybe reduce the folds if it takes too long.

    It is good to remark that cross validation is used only for measuring the model performance with a reasonable bias. Depending of whether you need to optimize hyperparameters or not, cross validation may not be needed in a production environment.

    Regards,
    Sebastian