https://github.com/alon-regev/ml_digits

https://github.com/alon-regev/ml_digits

Last synced: 24 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/alon-regev/ml_digits
• Owner: Alon-Regev
• Created: 2022-05-11T16:17:55.000Z (almost 3 years ago)
• Default Branch: master
• Last Pushed: 2022-05-24T15:40:28.000Z (almost 3 years ago)
• Last Synced: 2025-02-08T01:33:33.401Z (3 months ago)
• Language: Python
• Size: 11.8 MB
• Stars: 2
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Digit Classification

Simple project, learning about neural networks and supervised learning. Implemented neural network from scratch using a gradient descent algorithm. 



Project contains:

- A training program which produces a file of the trained model (takes a few minutes).

- An interactive digit classification which allows drawing in real time.


![image](https://user-images.githubusercontent.com/71284855/170075572-b1c9a86b-f173-4b48-8808-7498c522879f.png)

- Simple digit classification from a picture. (picture should have a white background, like a paper).


![image](https://user-images.githubusercontent.com/71284855/170075391-eee841fb-114d-489e-a95a-ed1bd0e21930.png)

# How to use

run Main.py from any of the files. 


requires numpy, pillow and mnist dataset library for training progra (or you can use the already trained model).

# Accuracy

![image](https://user-images.githubusercontent.com/71284855/170076161-437d7bb4-3c36-4cdc-a7c2-ee514c7920ca.png)




It has some issues reading from paper with some of the digits, so the overall accuracy from a picture is around 67%. 


![image](https://user-images.githubusercontent.com/71284855/170076333-5dffee4a-9b7e-4185-94e2-516ce7792ec0.png)