A snake game written in Python using the Pygame library, with implementations for different intelligent agents to play the game. Agents include a random agent, reflex agent, Q-learning agent, approximate Q-learning agent, and deep Q-learning agent.

Use Python 3.7 or above

matplotlib==3.5.1

numpy==1.22.3

pygame==2.1.2

torch==1.11.0

torchvision==0.12.0

Check requirements.txt for a full list of packages.

In order to play the game as a human player, run python3 userAgent.py.

-a {String}, --agent {String} (options: "random", "reflex", "exactq", "approxq", "dqn")

-n {Integer}, --num_runs {Integer}

-v, --verbose (Extra detail per test run)

-g, --graphics (Turn on graphics for testing)

-p, --plain (Show no score on graphics)

-j, --json (run settings from testSettings.json)

-f {Integer}, --framerate {Integer} (change the framerate of the game)

-ff, --fixed_food (Turn off random food spawning)

--load_dqn {Filename} (For dqn, specify which model to load)

--load_exactq {Filename} (For exact q agent, specify which model to load)

--load_approxq {Filename} (For appro q agent, specify which model to load)

Test all agents with graphics for 5 runs each (random food spawning):

python3 test.py -a random reflex exactq approxq -n 5 -g

Note: DQNAgent is not included in the above list because it has a small chance of getting stuck in an infinite loop. The game will timeout after many steps, but this is boring to watch with graphics turned on.

Test Exact-Q & DQN agent with graphics for 5 runs each (fixed food spawning):

python3 test.py -a exactq dqn -n 5 -g -ff --load_dqn models/DQN_10000_fixed.pth

Observation: Note how differently these agents act with fixed food spawning.

Test DQN agent with graphics for 4 runs (random food spawn)

python3 test.py -a dqn -n 1 -g

Note: This agent may get stuck in a loop, and if it does, it will run until the game times out.

Test Random Agent for 1000 runs:

python3 test.py -a random -n 1000

-a {String} (Agent to use)

-n {Integer} (Number of training episodes)

-t {Integer} (Number of test runs)

-g (Use graphics)

-v (Verbose output)

-l (Load checkpoint from pickle file)

-s (Save trained weights or qvalues)

-r (Random food spawn during testing)

--checkpoint {Filename} (Filename of checkpoint to load)

--save_filename {Filename} (Filename for saving weights)

Train Exact-Q Agent for 5000 runs:

python3 qLearningTrain.py -a q -n 5000 -t 20 -s --save_filename exactq.pkl

Note: Weights will be saved to SRC/

Train Approximate-Q Agent for 5000 runs:

python3 qLearningTrain.py -a approxq -n 5000 -t 20 -s --save_filename approxq.pkl

Note: Weights will be saved to SRC/

-e {Integer}, --episodes {Integer} (number of episodes to train for)

-g, --graphics (show graphics)

-s, --save (save model)

-r, --random_food (spawn food randomly)

-d, --record_data (record run data as csv file)

-l, --load (load saved model)

Train DQN Agent for 5000 runs with fixed food spawning:

python3 dqnTrain.py -e 5000 -s

Note: Model & stats files will be saved in models/