Core_Inputs

Related files: Input Formatter Output Formatter Reward Manager

In order to improve performance, we do not use all the values the game_tick_packet gives us, but we also generate some other inputs which can give the bot an advantage.

Input Formatter

Depending on the input_formatter being used, the data being presented to the model will be different.

• create_input_array is used to translate the inputs(observation) into an array understandable for the model
• transform_tensor change the shape of the input tensor(using pytorch with tensorflow backend)
• get_input_state_dimension return the shape the translated inputs have

Output Formatter

The output object has the controls given in the exact same format as they are outputted by get_bot_output

• format_model_output is used to translate the outputs(actions) into a object valid to be outputted by the agent
• create_array_for_training is used to translate the outputs(actions) into an array understandable for the model
• get_model_output_dimension return the shape that outputs have before being translated

Reward Managers

the reward is defined from observations, and actions taken.

• create_reward will define a reward from the input and output data