MAE

In this task, let's try to understand how to determine whether the MAE score we obtained signals a good or a bad model performance (to start somewhere, we'll keep the threshold simple).

Consider the following dataset (where $y$ is the ground truth label, and $\hat{y}$ is the predicted label):

1) Calculate the mean absolute value for this dataset (we will call this value $\text{MAE}_{\text{original}}$).

2) Then, calculate the median of the ground truth label.

3) Calculate the MAE for a case where all predictions are equal to the median of the ground truth (so the new dataset the $\hat{y}$ column will have the same value, the ground truth median) — we'll call this value $\text{MAE}_{\text{baseline}}$.

4) Compare $\text{MAE}_{\text{original}}$ and $\text{MAE}_{\text{baseline}}$. We'll say that $\text{MAE}_{\text{original}}$ is good if it equals $\text{MAE}_{\text{baseline}} - 0.2$ or less.

Your answer function looks like this:

$$\text{Answer}= \begin{cases} \text{MAE}_{\text{baseline}} + 1, & \text{if }\ \text{MAE}_{\text{original}}\, \text{is good} \\ \text{MAE}_{\text{baseline}}, & \text{otherwise} \end{cases}$$