Perceptron Algorithm: Implementation from Scratch

Project Overview

This project involves building a Single-Layer Perceptron—the fundamental building block of Artificial Neural Networks—entirely from scratch using Python and NumPy.

Instead of using high-level API calls (like keras.layers.Dense), this repository manually implements the mathematical logic for:

1. Weighted Sum Calculation (Dot Product).
2. Activation Function (Step Function).
3. Weight Update Rule (Rosenblatt's Learning Algorithm).

Mathematical Logic

The Perceptron makes a prediction based on linear combinations of input features:

• Activation: The model uses a Heaviside Step Function: output is 1 if $z \ge 0$, else -1.
• Learning Rule: Weights are updated iteratively based on the error: $$w := w + \Delta w$$ $$\Delta w = \eta \times (target - prediction) \times x_i$$ (Where $\eta$ is the learning rate)

Tech Stack

• Language: Python
• Core Logic: NumPy (for vectorization and efficiency)
• Data Processing: Pandas
• Dataset: Iris Dataset (Binary Classification: Iris-setosa vs. Others)

Code Structure

The implementation is encapsulated in a Perceptron class with the following methods:

• __init__: Initializes learning rate and epochs.
• fit(X, y): Trains the model by iterating through the dataset and updating weights.
• net_input(X): Calculates the dot product $w \cdot x + b$.
• predict(X): Applies the step function to return class labels.

How to Run

1. Clone the repository:

   git clone [https://github.com/PyPro2024/Perceptron-Algorithm-From-Scratch.git]

2. Install dependencies:

   pip install numpy pandas scikit-learn

3. Run the Notebook: Open Perceptron.ipynb in Jupyter Notebook or Google Colab to see the training process and accuracy results.

Results

• Training: The model successfully converges and learns to separate the linearly separable classes of the Iris dataset.
• Accuracy: Achieved 100% accuracy on the test set for this specific binary classification task.

---

If you find this foundational project helpful, feel free to ⭐ the repo!