This PR:

• in lesson 5:
  • runs the optimization with a fixed number of iterations, record the loss history, and use the parameters at the lowest validation loss as the solution.
  • uses $L$ instead of $\mathrm{loss}$
  • uses the means instead of the sums for both the loss and regularization term
  • decouples the calculations of the raw loss and the regularization
  • monitors the raw loss instead of the regularization loss
• adds lesson 7
• adds an appendix to explain terms

To see a cell by cell diff of lesson 5: Screenshot 2021-08-10 at 14-06-38 nbdime - diff and merge your Jupyter notebooks