Add proper evaluation for models

[skyfenton]

As we experiment with different additions, we should finally have a way to evaluate how well our models are doing. There are two metrics which seem relevant in this case:

• Precision at k, or the percent of recommended movies in the top k recommendations for a given user that were movies they liked (good recommendations). This metric does better the less movies we mark as good recommendations which aren't (false positives).
• Recall at k, or the percent of known positive movies in the top k recommendations for a given user that we also correctly recommend. This metric does better the more movies we pick that the user does indeed like (less false negatives).

We should consider more metrics as the project matures (ROC AUC, RMSE?).

[JamesKohlsRepo]

Sounds exactly like what we need, if you would like any help delivering part of this issue let us know