🧠 Machine Learning Models with PyTorch

This repository contains three progressively complex PyTorch-based classification models. Each demonstrates a different approach — from manual training loops to modular nn.Module structures — showcasing the evolution of model-building skills.

---

📂 Repository Navigation

Machine-Learning-Models/
│
├── 01_pytorch_workflow.ipynb       # Manual binary classification with raw tensors
├── 02_MCM.ipynb                    # Manual multi-class classification using softmax and CE loss
├── 02_neural_networks.ipynb        # nn.Module-based binary classifier using BCEWithLogitsLoss
└── README.md                       # Project documentation (this file)

---

🚀 Models Overview & Results

🔹 Model 1 — 01_pytorch_workflow.ipynb

• Type: Linear Regression Model
• Loss: Mean loss
• Optimizer: Manual Gradient Descent
• Final Accuracy: ~95%
• Final Loss: ~0.05

🔹 Model 2 — 02_MCM.ipynb

• Type: Multi-Class Classification (Manual)
• Loss: CrossEntropyLoss
• Optimizer: SGD
• Final Accuracy: ~92%
• Final Loss: ~0.18

🔹 Model 3 — computer_vision.ipynb

• Type: CNN model on image classification (Modular)
• Loss: BCEWithLogitsLoss
• Optimizer: SGD (lr=0.01)
• Final Accuracy: ~83%
• Final Loss: ~0.4

Note: Metrics may slightly vary depending on seed and hardware.

---

🧰 Features

• ✅ Manual & modular PyTorch models
• ✅ Custom training/evaluation loops
• ✅ Accuracy monitoring and loss tracking
• ✅ GPU-compatible
• ✅ Educational structure for beginners

---

🔧 Requirements

pip install torch torchvision matplotlib tqdm

---

📈 Planned Improvements

• Add training visualizations
• Export .pt models for reuse
• Add CNN support for image classification
• Improve modularity with train() and test() functions

---

👨‍💻 Author

Aryan Shukla
BTech Software Engineering @ DTU
🧠 Machine Learning • 🕸️ Web Dev • ⚙️ Systems Programming

---

⭐ Project Rating

🔍 Current Score: 8.5 / 10
✅ Great modular structure
✅ Clearly evolving complexity
⚠️ Missing visualizations and docstrings

Feel free to ⭐ star the repo or open a PR to contribute!

---

📬 Questions / Contributions

Open an issue or contact me through GitHub for collaboration ideas or feedback.