Neural Networks from Scratch (Python & NumPy)

This repository serves as a fundamental exploration into the mathematics of Deep Learning. It contains pure Python implementations of various Neural Network architectures—built without the aid of high-level frameworks like TensorFlow or PyTorch—to demonstrate the core algorithms driving modern AI.

Project Modules

1. Multi-Layer Perceptrons (MLP)

File: BasicNetwork.py A collection of dense network architectures demonstrating the evolution of learning capacity:

• Single Layer: Implements basic linear separation logic.
• Double Layer: Introduces hidden layers and non-linear activation functions (Sigmoid) to solve more complex patterns.
• Triple Layer: A deep network implementation featuring full backpropagation and error calculation.

2. Recurrent Neural Networks (RNN) & NLP

Files: TextNatureRNN.ipynb, data.py A custom RNN implementation designed for Natural Language Processing tasks, specifically sentiment analysis.

• Architecture: Manages hidden states (h) and time-step unrolling manually.
• Forward Pass: Uses Tanh activation for state updates and Softmax for output probabilities.
• Backpropagation Through Time (BPTT): Calculates gradients across time steps to learn sequence dependencies.
• Data: A custom labeled dataset of natural language phrases (e.g., "i am very happy") mapped to sentiment booleans.

3. The Neuron Primitive

File: Neuron.py The atomic unit of the network. A class-based implementation of a single neuron handling weights, biases, and feedforward calculations.

4. Training & Visualization

File: NeuralNet.ipynb A Jupyter Notebook environment used to visualize the training process. It includes:

• Real-time loss calculation (Mean Squared Error).
• Epoch-by-epoch accuracy tracking.
• Dynamic user input for inference testing.

---

Tech Stack

• Language: Python 3.x
• Core Math: NumPy (Matrix multiplication, Dot products, Exponentials)
• Visualization: Matplotlib / Pandas

How to Run

1. Install Dependencies:

   pip install numpy pandas matplotlib

2. Train the MLP Models:

   python BasicNetwork.py

3. Run the RNN Sentiment Analyzer: Open TextNatureRNN.ipynb in Jupyter Notebook or VS Code to train the model on the text dataset and see it predict sentiment on new sentences.

---

Built from first principles.