danielaleite / becooless Goto Github PK

Observe the unexpected behavior reported by @saksham36:

For “location”: “Rewe”, “shopping_list”: {“bananas”: 4, “beef”: 2, “carrot”:3}, the optimal price is {"price": 13.95, "carbon": 108.24}, but optimal CO2 is {"bio banana": {"amount": 4, "price": 3.0, "carbon": 36.0}, "beef": {"amount": 2, "price": 5.98, "carbon": 58.0}, "carrot": {"amount": 3, "price": 4.47, "carbon": 0.24}, "price": 13.45, "carbon": 94.24}.

In this particular setting, the CO2-optimized price is actually lower than the price-optimized price; clearly not what we want.

This is likely cause by the convergence criteria in the optimizer and can be improved by integrating one-hot constraints for the amount vectors for each category (for cheapest price, just buy the cheapest product in each category).

An outer-loop hotfix could be to run price and CO2 optimization, but if the price of CO2 is lower, then also use this solution as cheapest.