#learning-curve

Question 1: Single choice

You cannot collect more data, but you want evidence that more training data would lower your model's error. Which experiment gives that evidence?

Retrain on growing subsets and see if validation error still falls at full size1
Train longer on the full data and check whether the training error keeps going down2
Duplicate every training example to double the dataset, then compare validation errors3
Switch to a model with more parameters and check whether the training error drops further4