Are the hyperparameters to this recipe optimized?
Abhinav Sharma

Interesting idea! My evaluation metric would be, the average of 3-5 people:

  1. eat palette cleanser
  2. taste recipe
  3. rate on scale of 1–5

Issues include: normalizing scores for individuals; the slow decline into inebriation as you try more samples.

The base recipe from the Real Lemon bottle was for 1c lemon juice 1c sugar and 6 1/2 c water. Assuming that I wanted to hold lemon juice and liquid constant, I could tune the amounts of sugar, lavender water and gin, adding up to 6 1/2 c liquid with water for the extra. That gives three tunable parameters, requiring around 30 iterations for methods like SigOpt and Grid Search.