Deep Q learning play Tetris with interactive plots (WIP)

Requirements

$ mkdir venv
$ python -m venv venv/
$ pip install setuptools wheel
$ pip install matplotlib numpy pygame
$ pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

Train

    $ python main.py

Play

    $ python play.py

Remember to change fall_speed in tetris/constants.py to higher number.
n get new tile
r reset whole game
q/esc quit game

Experimented Combinations

States	Action Space	Tile Fall	Current Results
(cleared rows, holes, bumpiness, heights)	(up, down, right, left)	True	Fail
(cleared rows, holes, bumpiness, heights)	(up, right, left)	False	Fail
RGB frames	(up, down, right, left)	True	Fail
RGB frames	(up, right, left)	True	Fail
Binary grids	(up, down, right, left)	True	Fail
Binary grids	(up, right, left)	True	Fail

Experimenting

• Skip n-frames
• Weighted heights
• Simplified action space: (right, left, space)
• cols x 1 Convolution
• DDQN
• Skip Frames
• More algorithm