I have recently just completed a course on deep learning and I feel like an intermediate, but I still don't know how to structure this problem.
I'm looking to create a NN to play the card game Splinterlands.
I have the history of battles, the cards played, who won, and the battle rules and constraints.
I'm struggling with how best to model the inputs. I think that standard feature engineering and the encoding of all the variables affect the game, like manor, cards chosen, rules, etc.
How best to constrain the outputs? For example, the model needs to select cards I have available, etc., while the history contains cards I don't.
I know it's a long shot, but just looking for some inspiration :)
The basic rules of splinterlands are as follows:
- Start battle (you are given mana and set of rules, like no shields etc)
- Then select a summon (some rules say it has to be of certain, you also want to select your highest level summoner)
- Then select 6 monsters ( each monster has a position and stats like attack, speed, health and mana cost)
- Press battle
