For every day in the test data, we have an ideal weight vector that optimizes the gains for that day. Every predicted vector (w_p) is scored with an RMSE loss w.r.t. the optimal weight vector.

Submission with a lower total loss will be placed higher on the leaderboard.