1"""
  2# QuackNet
  3
  4**QuackNet** is a Python based building and training neural networks and convolutional networks entirely from scratch. It offers foundational implementations of key components such as forward propagation, backpropagation and optimisation algorithms, without relying on machine learning frameworks like TensorFlow or PyTorch
  5
  6## Key Features
  7
  8**1. Custom Implementation:**
  9-   Fully handwritten layers, activation functions and loss functions.
 10-   No reliance on external libraries for machine learning (except for numpy)
 11
 12**2. Core neural network functionality:**
 13-   Support for common activation functions (eg.Leaky ReLU, Sigmoid, Softmax)
 14-   Multiple loss functions with derivatives (eg. MSE, MAE, Cross entropy)
 15
 16**3. Training:**
 17-   includes backpropagation for gradient calculation and parameter updates
 18-   Optimisers: Gradient Descent, Stochastic Gradient Descent (SGD), and Adam optimiser.
 19-   Supports batching for efficient training.
 20
 21**4. Layer Support:**
 22-   Fully Connected Layer (Dense)
 23-   Convolutional
 24-   Pooling (Max and Average)
 25-   Global Average Pooling
 26-   Activation Layers
 27
 28**5. Additional Features:**
 29-   Save and load model weights and biases.
 30-   Evaluation metrics including accuracy and loss.
 31-   Visualisation tools for training progress.
 32-   Demo projects like MNIST and HAM10000 classification.
 33
 34## Installation
 35
 36QuackNet is simple to install via PyPI.
 37
 38**Install via PyPI**
 39
 40```
 41pip install QuackNet
 42```
 43
 44## Usage Example
 45
 46```Python
 47from quacknet.main import Network
 48
 49# Define a neural network architecture
 50n = Network(
 51    lossFunc = "cross entropy",
 52    learningRate = 0.01,
 53    optimisationFunc = "sgd", #stochastic gradient descent
 54)
 55n.addLayer(3, "relu") # Input layer
 56n.addLayer(2, "relu") # Hidden layer
 57n.addLayer(1, "softmax") # Output layer
 58n.createWeightsAndBiases()
 59
 60# Example data
 61inputData = [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]]
 62labels = [[1], [0]]
 63
 64# Train the network
 65accuracy, averageLoss = n.train(inputData, labels, epochs = 10)
 66
 67# Evaluate
 68print(f"Accuracy: {accuracy}%")
 69print(f"Average loss: {averageLoss}")
 70```
 71
 72## Examples
 73
 74-   [Simple Neural Network Example](/ExampleCode/NNExample.py): A basic neural network implementation demonstrating forward and backpropagation
 75-   [Convolutional Neural Network Example](/ExampleCode/CNNExample.py): Shows how to use the convolutional layers in the library
 76-   [MNIST Neural Network Example](/ExampleCode/MNISTExample/mnistExample.py): Shows how to use neural network to train on MNIST
 77
 78## Highlights
 79
 80-   **Custom Architectures:** Define and train neural networks with fully customisable architectures
 81-   **Optimisation Algorithms:** Includes Gradient Descent, Stochastic Gradient Descent and Adam optimiser for efficient training
 82-   **Loss and Activation Functions:** Prebuilt support for common loss and activation functions with the option to make your own
 83-   **Layer Support:**
 84    -   Fully Connected (Dense)
 85    -   Convolutional
 86    -   Pooling (max and Average)
 87    -   Global Average Pooling
 88    -   Activation layer
 89-   **Evaluation Tools:** Includes metrics for model evaluation such as accuracy and loss
 90-   **Save and Load:** Save weights and biases for reuse for further training
 91-   **Demo Projects:** Includes example implementations such as MNIST digit classification
 92
 93## Code structure
 94
 95### Neural Network Class
 96-   **Purpose** Handles fully connected layers for standard neural network
 97-   **Key Components:**
 98    -   Layers: Dense Layer
 99    -   Functions: Forward propagation, backpropagation
100    -   Optimisers: SGD, GD, GD using batching
101
102### Convolutional Neural Network Class
103-   **Purpose** Specialised for image data processing using convolutional layers
104-   **Key Components:**
105    -   Layers: Convolutional, pooling, dense and activation layers
106    -   Functions: Forward propagation, backpropagation, flattening, global average pooling
107    -   Optimisers: Adam optimiser, SGD, GD, GD using batching
108
109## Related Projects
110
111### Skin Lesion Detector
112
113A convolutional neural network (CNN) skin lesion classification model built with QuackNet, trained using the HAM10000 dataset. This model achieved 60.2% accuracy on a balanced validation set.
114
115You can explore the full project here:
116[Skin Lesion Detector Repository](https://github.com/SirQuackPng/skinLesionDetector)
117
118## License
119
120This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.
121"""