I am working on a project to try to train a network to play BTD6 (Bloons Tower Defense 6) Current my inputs are cash, lives, and the bloons in the next round. My outputs are the number of towers * a number of pre-determined spots on the map to place the tower.
However, I want to extend the functionality to support the dynamic placement of towers in the game (float x and y values) and support for upgrading towers. This would not be possible with a single network, I was looking to make a set of networks in a decision tree structure. I am having issues with how to implement this using this library.
Main Net - should output whether to buy a tower, upgrade a tower, or do nothing
If buy a tower was selected:
Sub Net 1 - should output what tower to build
Sub Net 2 - should output an x, y value (scaled to the height and width of placeable area)
If upgrade a tower was selected:
Sub Net 1 - should output what upgrade (0-0-1 0-0-2 0-1-0 ... etc)
Sub Net 2 - should output an x, y value to try to upgrade a tower
I am working on a project to try to train a network to play BTD6 (Bloons Tower Defense 6) Current my inputs are cash, lives, and the bloons in the next round. My outputs are the number of towers * a number of pre-determined spots on the map to place the tower.
However, I want to extend the functionality to support the dynamic placement of towers in the game (float x and y values) and support for upgrading towers. This would not be possible with a single network, I was looking to make a set of networks in a decision tree structure. I am having issues with how to implement this using this library.
Main Net - should output whether to buy a tower, upgrade a tower, or do nothing
If buy a tower was selected:
Sub Net 1 - should output what tower to build
Sub Net 2 - should output an x, y value (scaled to the height and width of placeable area)
If upgrade a tower was selected:
Sub Net 1 - should output what upgrade (0-0-1 0-0-2 0-1-0 ... etc)
Sub Net 2 - should output an x, y value to try to upgrade a tower