Ground-up implementations of ML algorithms

Overview

This project contains a curated collection of machine learning and deep learning algorithm implementations that are written from scratch (i.e., minimal dependencies), intended for learning, experimentation and research. The implementations are designed to work with popular frameworks like PyTorch and TensorFlow, along with GPU‑compatible training scripts and datasets.

Project Structure

Ground-up-implementations-ML-algorithms-/
├── algorithms/                         # Core algorithm implementations
│   ├── MLP_cifar_pytorch.py            # MLP using PyTorch
│   ├── MLP_mnist_tf.py                 # MLP using TensorFlow
│   ├── CNN_cifar_pytorch.py            # CNN using PyTorch
│   ├── CNN_mnist_tf.py                 # CNN using TensorFlow
├── dataset/                            # Dataset loader
├── notebooks/                          # Jupyter notebooks demonstration
│   ├── MLP_cifar_pytorch.ipynb         # MLP using PyTorch
│   ├── MLP_mnist_tf.ipynb              # MLP using TensorFlow
│   ├── CNN_cifar_pytorch.ipynb         # CNN using PyTorch
│   ├── CNN_mnist_tf.ipynb              # CNN using TensorFlow
│   ├── RNN_cifar_pytorch.ipynb         # RNN using PyTorch
│   ├── RNN_mnist_tf.ipynb              # RNN using TensorFlow
└── LLM_scratch.ipynb                   # LLM using PyTorch
├── LICENSE                             # MIT License
├── README.md                          

Featured Algorithms

#	Algorithm	Framework	Dataset	Code	Train Time	Accuracy
1	Multilayer Perceptron	PyTorch	CIFAR10	Github Colab	14 min	0.46
2	Multilayer Perceptron	Tensorflow	MNIST	Github Colab	22 s	0.96
3	Convolutional Neural Network	PyTorch	CIFAR10	Github Colab	22 min	0.63
4	Convolutional Neural Network	PyTorch	MNIST	Colab	2 min	0.96
5	Convolutional Neural Network	Tensorflow	MNIST	Github Colab	12 s	0.98
6	Recurrent Neural Network	From Scratch	Book by H.G. Wells "The Time Machine"	Colab	20 min	Perplexity = 1.2
7	Recurrent Neural Network	PyTorch	Book by H.G. Wells "The Time Machine"	Colab	3 min	Perplexity = 1.1
8	Large Language Model (Using Generative Pre-Trained Transformers)	PyTorch	Edith Wharton's "The Verdict"	Colab	28 s	Perplexity = 1.9

Code Usage

1. Clone the repository:

git clone https://github.com/rastri-dey/Ground-up-implementations-ML-algorithms-.git
cd Ground-up-implementations-ML-algorithms-

2. Install dependencies:

pip install -r requirements.txt

3. Run a model script or notebook:

python -m algorithms.<algorithm_script>

Note: Some models include Google Colab links to launch with minimal setup.

Contribution

We welcome contributions that:

• Improve algorithm correctness or efficiency
• Add new models or datasets
• Provide notebooks with visual explanations
• Improve documentation

Just open a Pull Request or Issue and we’ll help you get started!

License

This project is licensed under the MIT License.