Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/timhanewich/2048-ai

Training a sequential neural network to play the game 2048 using TensorFlow
https://github.com/timhanewich/2048-ai

Last synced: 2 months ago
JSON representation

Training a sequential neural network to play the game 2048 using TensorFlow

Awesome Lists containing this project

README 

          
# 2048 AI

I tried to train a sequential neural network to play the game 2048 using TensorFlow (keras). As seen below, this attempt seemed to be unsuccessful or not enough learning time was allocated.



## Exlaining the folders

- **models** - First implementation. Development ended on this commit *3cbd1287113f01dc7fc61f25e0850c716140733d*

- **models2** - Added more hidden layers, started new model from scratch.



## Structure 1 Tests

- 2023-01-29 04-35-08.354122

    - Avg Max Value: 94.72  

    - Avg Concentration: 17.55

- 2023-01-29 04-45-13.728451

    - Avg Max Value: 125.44  

    - Avg Concentration: 20.365

- 2023-01-29 21-26-03.641107

    - Avg Max Value: 124.16  

    - Avg Concentration: 23.3

- 2023-01-30 03-18-12.587640

    - Avg Max Value: 163.84  

    - Avg Concentration: 26.55

- 2023-01-30 13-56-54.905192

    - Avg Max Value: 181.76

    - Avg Concentration: 28.305

- 2023-01-31 04-08-31.118039

    - Avg Max Value: 126.72

    - Avg Concentration: 24.215



## Structure 2

I'm adding significantly more hidden neurons to a model and training from there. 



I halted it's training on January 31, 2023 @ 11:43 PM. It clearly is not working.



### Tests

- 2023-01-31 04-46-14.085862 (start/after only 1 training)

    - Avg Max Value: 105.4

    - Avg Concentration: 19.24375

- 2023-01-31 05-06-47.336688

    - Avg Max Value: 113.6

    - Avg Concentration: 21.1109375

- 2023-01-31 13-44-33.655633

    - Avg Max Value: 104.4

    - Avg Concentration: 16.6125

- 2023-01-31 15-45-44.891697

    - Avg Max Value: 82.4

    - Avg Concentration: 17.421875

- 2023-01-31 18-45-10.140012

    - Avg Max Value: 108.4

    - Avg Concentration: 18.45

- 2023-02-01 04-26-09.856992

    - Avg Max Value: 104.8

    - Avg Concentration: 18.059375



## models4

This uses a different learning method. This plays X number of games side by side. Then selects the game that worked out the best. Then trains the neural network on every decision that was made during that game.



**I did notice a critical error in my code around 9:50 AM EST on February 3, 2023. I was declaring the list properties of classes outside of the __init__ which is they they were building up so many child elements (shared amongst all instances of the class). I corrected it and continued on. Any model after 2023-02-03 12-41-39.969567 is trained using the new version**



### Tests

- 2023-02-01 20-06-10.626808

    - Avg Max Value: 128.4

    - Avg Concentration: 18.5890625

- 2023-02-01 23-52-37.388295

    - Avg Max Value: 135.2

    - Avg Concentration: 19.7078125

- 2023-02-02 02-46-35.077023

    - Avg Max Value: 125.8

    - Avg Concentration: 18.215625

- 2023-02-02 12-39-01.253694

    - Avg Max Value: 114.8

    - Avg Concentration: 17.4875

- 2023-02-02 12-39-01.253694

    - Avg Max Value: 126.8

    - Avg Concentration: 17.9609375

- 2023-02-02 16-42-30.178057

    - Avg Max Value: 136.0

    - Avg Concentration: 19.3109375

- 2023-02-03 04-19-27.234289

    - Avg Max Value: 167.0

    - Avg Concentration: 22.625

- 2023-02-03 12-41-39.969567

    - Avg Max Value: 148.8

    - Avg Concentration: 21.196875

- 2023-02-03 16-14-58.159941

    - Avg Max Value: 165.6

    - Avg Concentration: 22.9265625

- 2023-02-03 18-30-26.653947

    - Avg Max Value: 150.4

    - Avg Concentration: 21.30625

- 2023-02-05 21-25-17.395001

    - Avg Max Value: 170.4

    - Avg Concentration: 23.3328125

- 2023-02-05 22-53-50.155704

    - Avg Max Value: 164.4

    - Avg Concentration: 22.6171875

- 2023-02-06 18-18-31.158156

    - Avg Max Value: 172.0

    - Avg Concentration: 23.771875

- 2023-02-07 02-10-55.288370

    - Avg Max Value: 176.0

    - Avg Concentration: 24.2921875

- 2023-02-10 06-49-57.094073

    - Avg Max Value: 163.2

    - Avg Concentration: 21.8828125

- 2023-02-10 19-48-12.580502

    - Avg Max Value: 122.6

    - Avg Concentration: 18.0046875



## models5

I added more hidden neurons for this one. And it plays 10 games and chooses the best, not 20. I'm hoping this will allow it to train a bit faster (not spending so much time simulating a game).



### Tests

- 2023-02-12 19-31-40.975074

    - Avg Max Value: 155.8

    - Avg Concentration: 21.1671875

- 2023-02-12 21-36-06.716441

    - Avg Max Value: 126.8

    - Avg Concentration: 18.08125

- 2023-02-13 03-35-13.286957

    - Avg Max Value: 124.4

    - Avg Concentration: 18.875

- 2023-02-13 17-55-12.164313

    - Avg Max Value: 116.4

    - Avg Concentration: 17.61875

- 2023-02-14 12-42-35.487415

    - Avg Max Value: 142.0

    - Avg Concentration: 19.9578125

- 2023-02-14 19-46-22.543401

    - Avg Max Value: 130.8

    - Avg Concentration: 18.753125

- 2023-02-15 03-20-23.840451

    - Avg Max Value: 125.2

    - Avg Concentration: 18.4609375

- 2023-02-16 01-54-27.101415

    - Avg Max Value: 124.0

    - Avg Concentration: 18.534375