Let N be the number of examples, M be the number of features. We are trying to create a linear regression to fit the training samples.
- When N == M, do we have perfect 0 training loss?
- When N > M, do we have unique or many solutions?
- When N < M, do we have unique or many solutions?
Attempt:
- Depends, but if the instance is not colinear, then yes
- Yes, unique solution.
- Many solution.