KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #dsh-plugin #deepseek-harness #typescript #ai-agent #codex #developer-tools #claude-code #windows #mcp #python #desktop-app

litenet

★ 28 Open GitHub ↗

🛠️ A lightweight C++ neural network framework with 0 dependencies

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1 align="center">LiteNet</h1>

<p align="center">LiteNet is a neural network framework written in C++ with 0 external dependencies. It is designed to be easy to use, with a simple and intuitive API based on the Keras    and PyTorch libraries. All linear algebra operations are implemented from scratch with the C++ standard library.</p>

⚠ **This project is solely for educational purposes. It is not intended to be used in production.**

![LiteNet Model Training](https://raw.githubusercontent.com/martin226/litenet/main/docs/training.gif)

## Motivation

> Why make a neural network framework? Why in C++? Why no dependencies?

1. _Why make a neural network framework?_ I wanted to learn more about neural networks and how they work under the hood (backpropagation, optimization algorithms, etc.).
2. _Why in C++?_ I wanted to learn more about C++ and how to write efficient code. C++ is a powerful language that allows for low-level control and high performance.
3. _Why no dependencies?_ I wanted to challenge myself to implement everything from scratch. This includes linear algebra operations (matrix multiplication, element-wise operations, etc.), activation functions, loss functions, optimizers, etc.

## Example Usage (handwritten digit recognition)

For the full example, see [src/example_mnist.cpp](src/example_mnist.cpp).

Model architecture:

- Input layer (784 neurons)
- Dense layer (256 neurons, ReLU activation)
- Dense layer (128 neurons, ReLU activation)
- Dense layer (64 neurons, ReLU activation)
- Dense layer (32 neurons, ReLU activation)
- Output layer (10 neurons, Softmax activation)
- Loss function: Categorical Cross Entropy
- Optimizer: Adam

```cpp
// Build
const double learningRate = 0.0001;
litenet::Model model;
model.add(std::make_unique<litenet::layers::Dense>(784, 256, "relu", std::make_unique<litenet::initializers::HeUniform>()));
model.add(std::make_unique<litenet::layers::Dense>(256, 128, "relu", std::make_unique<litenet::initializers::HeUniform>()));
model.add(std::make_unique<litenet::layers::Dense>(128, 64, "relu", std::make_unique<litenet::initializers::HeUniform>()));
model.add(std::make_unique<litenet::layers::Dense>(64, 32, "relu", std::make_unique<litenet::initializers::HeUniform>()));
model.add(std::make_unique<litenet::layers::Dense>(32, 10, "softmax", std::make_unique<litenet::initializers::GlorotUniform>()));
model.compile("categorical_crossentropy", std::make_unique<litenet::optimizers::Adam>(learningRate));

// Train
const int epochs = 8;
const int batchSize = 128;
model.fit(trainingInputs, trainingTargets, epochs, batchSize, validationInputs, validationTargets);

// Predict
litenet::Matrix predictions = model.predict(inputs);

// Evaluate
std::vector<double> results = model.evaluate(testingInputs, testingTargets);
std::cout << "Loss: " << results[0] << std::endl;
std::cout << "Accuracy: " << results[1] << std::endl;
```

Results: (on MNIST dataset: 50,000 training samples, 10,000 validation samples, 10,000 testing samples)

Loss: 0.136428
Accuracy: 0.9364
```

## Features

- [ ] Layers
  - [x] Dense
  - [ ] Conv2D
  - [ ] MaxPooling2D
  - [ ] Flatten
  - [x] Dropout
- [x] Activation Functions
  - [x] ReLU
  - [x] Leaky ReLU
  - [x] Sigmoid
  - [x] Tanh
  - [x] Softmax
- [x] Loss Functions
  - [x] Mean Squared Error
  - [x] Mean Absolute Error
  - [x] Binary Cross Entropy
  - [x] Categorical Cross Entropy
- [x] Optimizers
  - [x] Stochastic Gradient Descent (mini-batch)
  - [x] Adam
  - [x] AdamW
  - [x] RMSprop
  - [x] Adagrad
- [ ] Metrics
  - [x] Accuracy
  - [ ] Precision
  - [ ] Recall
  - [ ] F1 Score
- [x] Initializers
  - [x] Random Normal
  - [x] Random Uniform
  - [x] Glorot Normal
  - [x] Glorot Uniform
  - [x] He Normal
  - [x] He Uniform
  - [x] Zeros
  - [x] Ones
- [ ] Regularizers
  - [ ] L1
  - [ ] L2

litenet

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README