MNIST Digit Recognition

Based on handwritten digits - which a sampled is displayed hereinabove - stored in the MNIST database (/res/...) and parsed by the great and useful toolbox mnist written by Marc Garcia. This program uses feedforward and backpropagation in a 3-layers Neural Network seen below to learn how to recognize the former mentionned digits :

Done

• Feedforward
• Backpropagation
• Gradient Descent
• Fit
• Predict
• Accuracy Test

To Do

• Customizable number of layers
• Cool terminal-based UI
• Continuous Integration (GitHub Actions, Pytest)
• Use none-MNIST handwritten digit as inputs
• Any more ideas ? Open an Issue !

Loss and Accuracy

For a batch size of one, five epochs and a learning rate of 0.01, we get these graphs :

Setup

To setup the project on your local machine:

1. Click on Fork.
2. Go to your fork and clone the project to your local machine.
3. git clone https://github.com/master-coro/mnist-digit-recognition
4. pip3 install -r requirements.txt

Run

To run the project:

1. Cd into the root of the project cd path/mnist-digit-recognition
2. Run the main script with necessary args : python src/main.py

Contribute

To contribute to the project:

1. Choose any open issue from here.
2. Comment on the issue: Can I work on this? and get assigned.
3. Make changes to your fork and send a PR.

To create a PR:

Follow the given link to make a successful and valid PR: https://help.github.com/articles/creating-a-pull-request/

To send a PR, follow these rules carefully,otherwise your PR will be closed:

1. Make PR title in this format: Fixes #IssueNo : Name of Issue

For any doubts related to the issues, i.e., to understand the issue better etc, comment down your queries on the respective