This lab contains an example of two models(code block just before 7-Iterate to find optimal regularisation value), in which one model’s performance in the training set is better than the other model, but in the case of the dev set, the exact opposite is observed. Which model is ‘better’?
Really what you care about is the test set performance, because it represents fresh examples that were not used in either training or optimizing the model.
2 Likes