bluedi / deepdip Goto Github PK

Compare the results of several matches between the bots to prove the algorithm efficiency.

Implement a DeepDip v0 that uses DQN to play Diplomacy.

Create a map with fewer provinces, fewer borderlands, and fewer players.

Improve DeepDip to its final version, v1.

This should be achieved by implementing known improvements to the DQN algorithm such as Dueling Networks, or Prioritized Experience Replay.

After improving DeepDip's algorithm to v1, compare again against other bots to show the results of the improvements.

Use DumbBot to generate training data for DeepDip to use in the training process.

This includes the process of creating the format for the data that DumbBot will output, and that DeepDip will receive has input.

Implement a version of DumbBot in the BANDANA framework to better understand how to handle the framework and its inputs and outputs.

In order to get acquainted with the BANDANA framework, some experiments will be done.

• Study customization options;
• Try a public bot.