Notes:

• Kun doesn’t agree with recommendation system necessarily, can’t think of a use case for it

  • But does agree with (Matrix completion, and Collaborative filtering) and is open to expanding to other methods later on 

  • “Is there a way for one complicated model to output our outcomes, or will it have to be multiple simpler algorithms wrapped into one”? 

• K-folds: 

  • “it’s more of a multiple-choice than a single value”

  • Threshold: 

    • Will vary based on the number of potential outcomes

      • Anything better than random guess is good, e.g. if 2 outputs, anything better than 50% is better

    • Fairly low risk to accept the model’s output, since someone still has to confirm it