Why we first use dev set to find bias or variance problem?Why we don’t use training set?
Because the training set only shows how well the model has fit the data it was trained on. A model can have low training error but act poorly on the dev set (overfitting (high variance)). Conversely, poor performance on both sets suggests underfitting (high bias). The dev set helps us evaluate how the model generalizes.
Hope it helps! Feel free to ask if you need further assistance.
1 Like