https://github.com/antonlydike/fast2048

A try at creating the fastes posible implementation of the game 2048 in python
https://github.com/antonlydike/fast2048

Last synced: 23 days ago
JSON representation

A try at creating the fastes posible implementation of the game 2048 in python

• Host: GitHub
• URL: https://github.com/antonlydike/fast2048
• Owner: AntonLydike
• Created: 2021-03-29T20:49:25.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2021-03-30T10:43:50.000Z (almost 4 years ago)
• Last Synced: 2024-10-30T11:49:31.601Z (2 months ago)
• Language: Python
• Size: 5.86 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Fast 2048 in python

Trying to implement the fastest 2048 simulation in python, in order to use it as a machine learning base.

Currently the basic game is JIT compiled using numba. These are the performance characteristics (time per single run):

|run      | JITed  | Pure Python |

|---------|--------|-------------|

|1        | 3.4s   | 3.3ms       |

|2-1001   | 0.13ms | 2.7ms       |

|1002-2001| 0.13ms | 2.7ms       |

As you can see, the first run of the JITed version takes a considerable length of time, but afterwards the performance is just stellar (only requires 5% of the time). This is a twenty-fold improvement.

This means, that after approximately 1350 runs, the JITed version has broken even, and after 10000 iterations, about 22 seconds will be saved (the JITed version will take just about 4.7s, while the pure python one takes 27s).

# Q-Learning

The file `ml.py` holds a very rough-around-the-edges q-learning approach. I've developed this code during a course at my uni, it is far from perfect. In order to make full use of numba, and allow for full multiprocessing later on, I removed all classes and am just operating on numba types such as dicts and lists. This code, compared to the old, performs around 30 times faster. The initial run including compilation is about as quick as one run in standard python (both around 8 seconds).