31st (about 15th before I broke my bot, being a tester)
I have a Monte Carlo search, which is quite similar to the one I wrote for Code4Life.
I generate 5 random actions. I hardcoded testing 500 plans (my sim is fast enough to get some more). An action is a USE here, so it can make me move accross the whole map. An action has to be valid, so the options to choose from depend on the item, that my chef currently has (if any).
My chef is not a teamplayer at all. I mostly ignore my partner,
After each of my 5 random actions I score 2 things:
-
preparing the more complex foods (chopped strawberries, croissant, tart)
double timeWeight = 1 / Math.Log (2 + TravelTime); double factor = Math.Sqrt (Math.Min (1, itemCount / itemNeeded)) - Math.Sqrt (Math.Min (1, previousItemCount / itemNeeded)); Score += factor * itemCost * timeWeight; -
serving a dish to a customer
Score += 3 * SearchDepth * customer.Award * timeWeight;
This gives less score for finishing a task later (in terms of the real game time, not search depth for my actions - so it’s fine to pick up a dish before a save the tart from burning.
I have a higher motivation to get my 1st croissant, than a 3rd tart.
Taking extra time gives a slightly smaller score.
When I found my best plan, I replace USE by MOVE, if I’m not next to my target already. That’s the only time, that I check the postion of my partner. I try to get close to the next cell in my list in this stage, so that I might save a turn later.
When I place an item on a table, I prefer central tables with many neighbors over corners. That makes it easier to pick up an item again, it’s easier to reach.
I ended up with a little less than 1000 lines of code. This is mostly the game logic itself, a little path finding. The actual bot is pretty simple.