gym-game2048

rl-video-episode-8.mp4

Install

git clone https://github.com/helpingstar/gym-game2048.git
cd gym-game2048
pip install -e .

Example

import gymnasium as gym
import gym_game2048

env = gym.make("gym_game2048/Game2048-v0", render_mode="human")

observation, info = env.reset(seed=42)
for _ in range(1000):
    action = env.action_space.sample()  # this is where you would insert your policy
    observation, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
       observation, info = env.reset()
env.close()

Description

A reinforcement learning environment based on the game 2048.

The 2048 game is a single-player puzzle game where the objective is to combine tiles with the same number to create a tile with the value 2048. Tiles can be moved up, down, left, or right, and when two tiles with the same number touch, they merge into one tile with the combined value. The game ends when there are no more moves available, or when a tile with the value 2048 is created.

Action Space

There are 4 discrete deterministic actions:

0 : Swipe left
1 : Swipe right
2 : Swipe up
3 : Swipe down

Observation Space

The observation is a ndarray with shape (1, size, size).

The elements correspond to the following:

0 : Blank cell
1~ $\log_{2} (\text{goal})$ : $2^n$

Rewards

If you reach the goal, you get a reward of 1; if you don't reach the goal and the game ends, you get a reward of -1. In all other cases, you get a reward of 0.

If you want to change the reward policy, use Wrappers.

Arguments

size : The size of the board. The board will be made to be the size of (size, size).
goal
- The number you want to reach. It should be entered as a power of two. The game ends when one of the numbers on the board becomes the goal.
- This environment supports up to $2^{255}$. However, it is difficult to reach even $2^{20}$, so it is not recommended to enter more than this as a factor.

Wrappers

This is a personally created Wrapper. When applying, pay attention to the priority of wrappers.

Normalize2048

Divide the number on the board by the log of goal, which is the largest number that can be on the board.

Example)

# goal == 2048, log of goal == 11
# Render
| 512 | 16 | 4 | 8 |
# Original Observation
| 9 | 4 | 2 | 3
# Normalized Observation
| 0.81.. | 0.36.. | 0.18.. | 0.27.. |

RewardConverter.

Set the reward system in the same way as the scoring system in 2048.

div_pos_rew=2048 : This is the number that will be divided when points are scored.
default_neg=0 : This is the reward for not scoring points.
term_rew=-1 : This is the reward received at the end of the game.

TerminateIllegal

The episode ends immediately upon Illegal Action.

illegal_reward : Reward to receive when terminated due to illegal action