DLS C3 W1 Bird Recognition in Peacetopia Q11

I am not able to understand the case here. Please help me get this one…

How is it overfit (low basis and high variance). If we tentatively trace the calculation steps, what could likely have caused it?

And why is it not overfitting on train set? If given the choice, will we add data to the train set or dev set or both; in what manner?

1 Like

Hey @Shashwat_Gupta and welcome to the DLS,

We can’t discuss right or wrong answers on the quiz here. That said, we can discuss general cases and bias and variance tradeoff. Please, check this post out. I believe you may find answers on your questions there.

Note that we use a train set for training only, dev and test sets for evaluation of our models.

The choice would depend on a particular case and our intention.

  • We may add data to the train set to address variance on dev and test sets.
  • We may add data to the dev set to address data mismatch in dev and test sets.

@Shashwat_Gupta have you found answers on your questions? Please, feel welcome to ask further questions.