You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
All currently implemented player encodings are destined to fail.
The current encoding does not accommodate for any games in which the AI agent is not player one since all encodings are in player order, i.e., player one, player two, and so on. And since there is no way for the agent to know which player it represents, the agent would likely fail to understand the encoding.
Therefore, a new encoding system is proposed, which always puts the player whose turn it is in the first position of the encoding. Additionally, other players are simply referred to as opponent one to three, in order.
The new encoding would, therefore, keep the player the AI represents always in the same position in the vectorized state for each agent, whether it is player one or any of the others. This, in turn, would provide a more generalized and streamlined model input.
old
new
player one
current player
player two
opponent one
player three
opponent two
player four
opponent three
The text was updated successfully, but these errors were encountered:
All currently implemented player encodings are destined to fail.
The current encoding does not accommodate for any games in which the AI agent is not player one since all encodings are in player order, i.e., player one, player two, and so on. And since there is no way for the agent to know which player it represents, the agent would likely fail to understand the encoding.
Therefore, a new encoding system is proposed, which always puts the player whose turn it is in the first position of the encoding. Additionally, other players are simply referred to as opponent one to three, in order.
The new encoding would, therefore, keep the player the AI represents always in the same position in the vectorized state for each agent, whether it is player one or any of the others. This, in turn, would provide a more generalized and streamlined model input.
The text was updated successfully, but these errors were encountered: