Case context: You are training a model to predict monthly apartment rent in Seattle from apartment size in square feet. Your main dataset is limited, so you consider adding a much larger dataset of apartment rents from Phoenix. For apartments of the same size, rents in Phoenix are typically much lower than in Seattle.

Question: Should the Phoenix apartment dataset be added to the training set? Explain your answer using the idea of data consistency.

Sample answer: No. The Phoenix dataset should be excluded because it is not a consistent auxiliary source for the Seattle rent prediction task. The same input feature, apartment size, is associated with different target values in the two cities. Since the relationship between size and rent changes substantially by location, combining the datasets would teach the model conflicting patterns and reduce its usefulness on Seattle rent predictions.

Key points:

• Do not add the Phoenix data to the training set.
• The auxiliary data is inconsistent with the Seattle data.
• The same feature value maps to different rent levels across the two cities.
• Mixing the two sources would likely hurt performance on the Seattle task.

Rubric: The answer must clearly recommend leaving out the Phoenix data and explain that it is inconsistent because the mapping from apartment size to rent differs across the two locations.