bensoltoff
commented
6 years ago Use the mean-squared error. That will be both the loss function and the validation metric. This is fine because the model is learning using the training MSE, whereas you are evaluating the model's performance using the validation set MSE, and finally the test set MSE.
The second part asks us to report the test mean square errors of models. But the example of the Boston housing data in the textbook and the slide use mean absolute errors for evaluation. I'm wondering which we should use in the problem set, the mse from the loss function or the mae from the metrics.