https://github.com/parxd/groundupneuralnet
C++ neural network library w/ support for various activation & loss functions
https://github.com/parxd/groundupneuralnet
deep-learning linear-algebra neural-network neural-networks
Last synced: 13 days ago
JSON representation
C++ neural network library w/ support for various activation & loss functions
- Host: GitHub
- URL: https://github.com/parxd/groundupneuralnet
- Owner: Parxd
- License: mit
- Created: 2023-01-10T09:02:56.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2023-05-22T04:24:10.000Z (about 3 years ago)
- Last Synced: 2025-02-24T06:31:50.947Z (over 1 year ago)
- Topics: deep-learning, linear-algebra, neural-network, neural-networks
- Language: C++
- Homepage:
- Size: 4.54 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# GroundUpNeuralNet
GroundUpNeuralNet is a fast & efficient C++ library for constructing (deep) neural networks.
This library was built with [Eigen 3.4.0](https://eigen.tuxfamily.org/index.php?title=Main_Page) for its powerful vectorization capabilities and [GoogleTest](https://github.com/google/googletest) for unit testing.
**NOTE:** these dependencies come w/ this project's CMake, so no manual linking needed
It also comes with sample data generators to test your network (i.e. sine wave with noise), a trainer class for easier training using mini-batch optimization, and a save/load feature to save your model's weights and biases.
The following activation functions are supported:
- Sigmoid
- ReLU
- LeakyReLU
- Hyperbolic tangent
- Softmax
The following loss functions are supported:
- Mean squared error
- Categorical cross-entropy
As an example, a neural network can be constructed following this pattern:
```cpp
Container container(
new Linear(2, 10),
new ReLU,
new Linear(10, 10),
new ReLU,
new Linear(10, 2),
new Softmax
);
// ...or using BaseModule's factory methods
Container container(
BaseModule::make(784, 15),
BaseModule::make(),
BaseModule::make(15, 10),
BaseModule::make(),
BaseModule::make(10, 5),
BaseModule::make()
);
```
Another example: Using the sample data generator, trainer class (w/ cross-entropy loss), and save feature:
```cpp
auto data = Sine::generate(1000000, 0.2, 200, 50, 1.1);
Eigen::MatrixXf features = data.topRows(2);
Eigen::MatrixXf labels = data.bottomRows(2);
Trainer::train(cont, features, labels, 32, 5, 1, true);
cont.save("../src/model.csv");
```
An example run using sample data (noisy sine wave approximation, batch size of 32):
Future features I’d like to implement:
- Support for generic numeric types, such as doubles & integers. As of now, only float32 is supported.
- Support for different optimizers
- GPU support w/ CUDA