🌼 Iris Neural Network: From Scratch vs. PyTorch

This project showcases how to build a neural network from scratch using NumPy and compare it with an equivalent model implemented using PyTorch. The focus is on learning, understanding, and visualizing how neural networks learn to classify real-world data.

I used the Iris dataset, which is a classic multi-class classification problem involving three flower species.

---

📚 Overview

• 🧠 Manual Neural Network built with NumPy
• ⚙️ Equivalent model built using PyTorch
• 📉 Loss curves for both models
• 🎯 Prediction visualization and comparison
• 🌈 Decision boundary plot to show model behavior

---

🗂️ Project Structure

├── iris_nn.py # NumPy implementation ├── iris_nn_pytorch.py # PyTorch implementation ├── plot_decision_boundaries.py # For visualization ├── iris_nn_comparisonIris Neural Metwork From Scratch to PyTorch.ipynb # Jupyter notebook version ├── loss_manual.npy # Saved manual loss curve ├── loss_pytorch.npy # Saved PyTorch loss curve ├── plot_loses.png # Combined loss plot ├── plot_des_bon.png # Decision boundary plot ├── scratch_predictions.npy # Predictions by manual model ├── pytorch_predictions.npy # Predictions by PyTorch model ├── torch_model.pth # Serialized PyTorch model state └── README.md

---

🌸 The Iris Dataset

• Features:
  • Sepal length
  • Sepal width
  • Petal length
  • Petal width
• Classes:
  • Setosa
  • Versicolor
  • Virginica

---

🔧 Neural Network Architectures

✅ NumPy (From Scratch)

• 4 input neurons → 6 hidden (ReLU) → 3 output (Softmax)
• Manual implementation of forward/backward propagation
• Cross-entropy loss and basic gradient descent

⚡ PyTorch

• Same architecture using torch.nn.Linear
• Uses ReLU activations, softmax, and torch.optim.Adam
• Faster training and optimized performance

---

📈 Training Performance

🧪 Loss Curves

Both models successfully minimize the loss. PyTorch converges more quickly due to advanced optimizers, while the manual model shows a steady decline, proving the custom backpropagation logic works.

---

🧠 Decision Boundary

The decision boundary visualizes how each model separates classes using only two selected features. Both models show strong generalization and class separation.

---

🚀 Running the Project

✅ Dependencies

• Python 3.8+
• NumPy
• Matplotlib
• scikit-learn
• PyTorch
• Jupyter Notebook (optional)

🛠️ Setup

pip install numpy matplotlib scikit-learn torch notebook

▶️ Steps

• Train the from-scratch model

python iris_nn.py

• Train the PyTorch model

python iris_nn_pytprch.py

• Generate plots

python plot_decision_boundaries.py

• Explore in Jupyter Notebook

jupyter notebook Iris Neural Network From Scratch to PyTorch.ipynb

---

💡 Key Learnings

• How neural networks work under the hood

• Manual backpropagation and gradient updates

• Benefits of using deep learning frameworks like PyTorch

• Visualizing predictions and decision boundaries

---

👤 Author

M26I

© 2025 M26I – For educational/portfolio use only.
Unauthorized use or redistribution without credit is prohibited.