Tetris AI bot using custom DQN (Deep Q-learning network)

javac -cp "./lib/*;." @tetris.srcs
java -cp "./lib/*;." edu.bu.tetris.Main

There are several command line options available to you. If you want to see the exhausive list, please add a -h or --help argument to the command line and see the help message! Most of these command line arguments are your way of configuring the way you train your model (such as learning rate, batch size, etc.).

The hyperparameters we ended up going with

java -cp "./lib/*;." edu.bu.tetris.Main -q src.pas.tetris.agents.TetrisQAgent -p 10000 -t 400 -v 200 -b 10000 -n 0.001 -g 0.99 -c 10000000 -u 5 -s | tee run.log 

change the iteration to the latest iteration model trained

java -cp "./lib/*;." edu.bu.tetris.Main -a src.pas.tetris.agents.TetrisQAgent -i ./params/qFunction<iteration>.model

src/pas/tetris/agents/TetrisQAgent.java

Feed Forward Neural Network. 1 input layer, 1 hidden layer, 1 output layer

Input layer 37 nodes

• Height of each column
• Closest hole from top for each column
• Number of hole following it for each column
• Next 3 minos types
• Number of lines cleared for this move
• Total holes below the maximum height
• Tetris for this move
• T spin for this move

Hidden layer

• 64 nodes

Output layer

• 1 node representing the q value of this move

Reward function: -0.51 * height - 0.36 * holes - 0.18 * bumpiness + 7 * game.getScoreThisTurn();