Consider changing from MSE to MAE

[evantey14]

Parity scorers currently use mean squared error (MSE) on a target value to evaluate scores since that seemed the more natural error to use. However, we want an empty league to have a meaningful score, so I've normalized the MSE (otherwise the scores would just be zeros until there are at least a large # of kids on each team, meaning assignment wouldn't actually use the score in any decisionmaking). This normalization means that at the end of the process, the error is very very small so all the scores look like 1s.

I think this means we're making more trades in the early / mid-game but less in the late game because the scores are used linearly to calculate the final score.

Other options in general seem to be:

• picking a different normalization so we are more effective late game
• using a different error (like MAE)
• have a running target value based on currently assigned players (I believe this would be pretty slow to compute)
• have the normalization be calculated on construction (normalization depends on already assigned players from the input file)