Ecosyste.ms: Awesome

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

Awesome Lists | Featured Topics | Projects

https://github.com/farid-karimi/digit-recognizer

Digit Recognizer Neural Network
https://github.com/farid-karimi/digit-recognizer

digit-recognizer matplotlib mnist neural-network numpy pandas

Last synced: 28 days ago
JSON representation

Digit Recognizer Neural Network

Awesome Lists containing this project

README 

        
# Digit-Recognizer

This is a Digit Recognizer Neural Network using only Pandas and NumPy on the [MNIST](https://www.kaggle.com/competitions/digit-recognizer/data) Dataset



## Project Goals

The goal of this project was to gain a deeper understanding of neural networks and their inner workings. Initially, the complexity of concepts like backpropagation, activation functions, and gradient descent seemed overwhelming. To overcome this, I decided to implement a neural network from scratch without using any libraries or wrappers. This hands-on approach allowed me to grasp the foundational concepts and gain a solid understanding of the underlying principles.



Fortunately, there were others who had undertaken similar projects, and I leveraged their experiences and resources available on the internet to build my own version.



## Learnings

Throughout this project, I gained knowledge in several key areas, including:



- Backpropagation: I learned how to propagate errors backward through the network, updating weights and biases to improve the network's performance.

- Activation Functions: I explored different activation functions and their significance in introducing non-linearity to the network, enabling it to learn complex relationships.

- Gradient Descent Techniques: I studied various techniques in gradient descent, such as batch gradient descent, to optimize the network's learning process.

- Setting a Good Learning Rate (Alpha): I discovered how to choose an appropriate alpha value for the gradient descent function, ensuring efficient convergence.



These were just a few of the many small but important details I encountered during this project, which greatly enhanced my understanding of neural networks.



## Project Timeline

The entire project took approximately 2-3 days. During this time, I devoted myself to learning the necessary concepts and implementing the neural network from scratch. The process was enjoyable and rewarding, as it allowed me to gain hands-on experience and deepen my knowledge.



## Resources

I would like to share the resources that proved invaluable in my journey of understanding neural networks:



- [But what is a neural network? | Chapter 1, Deep learning - YouTube](https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi)

- [Building a neural network FROM SCRATCH (no Tensorflow/Pytorch, just numpy & math) - YouTube](https://www.youtube.com/watch?v=w8yWXqWQYmU&t=26s)

- [Neural Networks Explained from Scratch using Python - YouTube](https://www.youtube.com/watch?v=9RN2Wr8xvro&t=55s)

- [How to Create a Neural Network (and Train it to Identify Doodles) - YouTube](https://www.youtube.com/watch?v=hfMk-kjRv4c&t=1036s)

- [Learning rate alpha in gradient descent | by Thamarasee Jeewandara | Medium](https://thamarasee.medium.com/learning-rate-alpha-in-gradient-descent-6cc6d7b6df43)

- [Simple MNIST NN from scratch (numpy, no TF/Keras) | Kaggle](https://www.kaggle.com/code/wwsalmon/simple-mnist-nn-from-scratch-numpy-no-tf-keras/notebook#Simple-MNIST-NN-from-scratch)



These resources provided valuable insights and guidance throughout my learning process, and I hope they prove helpful to you as well.