A PyTorch-like deep learning framework implementation for educational purposes. This project implements a complete neural network library with automatic differentiation, common layer types, optimizers, and example implementations.
- Tensor operations with automatic differentiation
- Neural network modules (Linear, ReLU, etc.)
- Optimizers (SGD, Adam)
- MNIST classification example with 97.16% accuracy
- Comprehensive evaluation metrics
- Clone the repository:
git clone https://github.com/nnayz/MyTorch.git
cd MyTorch- Install uv (if not already installed):
curl -LsSf https://astral.sh/uv/install.sh | sh- Create and activate a virtual environment with dependencies:
uv venv
source .venv/bin/activate # On Windows use: .venv\Scripts\activate
uv pip install .- Download MNIST Dataset: The MNIST dataset files are not included in the repository due to their size. You'll need to:
a. Download the CSV files from Kaggle: MNIST CSV Format
b. Place the files in the archive directory:
archive/mnist_train.csv(for training data)archive/mnist_test.csv(for test data)
-
Tensor Operations (
mytorch/__init__.py)- Basic arithmetic operations (+, -, *, /)
- Matrix operations (dot product, transpose)
- Automatic differentiation with backward pass
- Gradient computation and accumulation
-
Neural Network Modules (
mytorch/nn.py)- Linear layer with weights and biases
- Activation functions (ReLU)
- Loss functions (Cross Entropy)
- Forward and backward propagation
-
Optimizers
- SGD (Stochastic Gradient Descent)
- Adam optimizer implementation
The framework follows a modular design similar to PyTorch:
MyTorch/
├── mytorch/
│ ├── __init__.py # Core tensor operations
│ └── nn.py # Neural network modules
├── mnist_example.py # Training script
├── evaluate_mnist.py # Evaluation script
└── pyproject.toml # Project configuration and dependencies
Run the training script:
python mnist_example.pyThe script will:
- Load and preprocess MNIST data
- Create a neural network model
- Train for specified epochs
- Save model weights
Run the evaluation script:
python evaluate_mnist.py- Accuracy: 97.16%
- Macro Precision: 97.23%
- Macro Recall: 97.14%
- Macro F1-score: 97.15%
| Digit | Precision | Recall | F1-Score |
|---|---|---|---|
| 0 | 98.28% | 98.98% | 98.63% |
| 1 | 97.84% | 99.56% | 98.69% |
| 2 | 98.87% | 93.60% | 96.17% |
| 3 | 90.64% | 98.81% | 94.55% |
| 4 | 97.95% | 97.45% | 97.70% |
| 5 | 96.26% | 98.09% | 97.17% |
| 6 | 98.11% | 97.49% | 97.80% |
| 7 | 97.86% | 97.67% | 97.76% |
| 8 | 98.28% | 93.63% | 95.90% |
| 9 | 98.18% | 96.13% | 97.15% |
[[ 970 1 0 2 0 0 4 0 2 1]
[ 0 1130 0 3 0 1 1 0 0 0]
[ 3 7 966 39 3 0 2 6 6 0]
[ 0 0 3 998 0 6 0 0 1 2]
[ 1 1 1 1 957 0 5 6 1 9]
[ 2 0 0 9 2 875 1 1 1 1]
[ 3 3 0 2 4 11 934 0 1 0]
[ 2 5 6 5 0 0 0 1004 2 4]
[ 4 3 1 29 6 12 3 3 912 1]
[ 2 5 0 13 5 4 2 6 2 970]]
-
Strong Overall Performance
- 97.16% accuracy on test set
- Consistent performance across classes
- High precision and recall for most digits
-
Per-class Analysis
- Best performance on digit 1 (F1: 98.69%)
- Most challenging: digit 3 (F1: 94.55%)
- Very high precision across all classes (>90%)
-
Error Patterns
- Most confusion between visually similar digits (2↔3, 4↔9)
- Minimal confusion between dissimilar digits
- Balanced error distribution
Feel free to open issues or submit pull requests for improvements or bug fixes. Areas for potential enhancement:
- Additional layer types (Conv2D, MaxPool, etc.)
- More optimization algorithms
- Data augmentation techniques
- Support for other datasets
- Performance optimizations
MIT License - feel free to use this code for educational purposes.