Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/harujoh/KelpNet
Pure C# machine learning framework
https://github.com/harujoh/KelpNet
csharp deep-learning dotnet gpu machine-learning neural-network onnx opencl
Last synced: about 1 month ago
JSON representation
Pure C# machine learning framework
- Host: GitHub
- URL: https://github.com/harujoh/KelpNet
- Owner: harujoh
- License: apache-2.0
- Created: 2016-07-21T06:56:18.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2023-11-30T06:49:47.000Z (about 1 year ago)
- Last Synced: 2024-08-03T13:14:21.837Z (5 months ago)
- Topics: csharp, deep-learning, dotnet, gpu, machine-learning, neural-network, onnx, opencl
- Language: C#
- Homepage:
- Size: 17.3 MB
- Stars: 242
- Watchers: 20
- Forks: 28
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-dotnet-machine-learning - KelpNet
README
# KelpNet : Pure C# machine learning framework
[](https://opensource.org/licenses/Apache-2.0) [](https://ci.appveyor.com/project/harujoh/kelpnet) [](https://codecov.io/gh/harujoh/KelpNet)
```csharp
/* SampleCode */
FunctionStack nn = new FunctionStack(
new Convolution2D(1, 32, 5, pad: 2, name: "l1 Conv2D"),
new ReLU(name: "l1 ReLU"),
new MaxPooling(2, 2, name: "l1 MaxPooling"),
new Convolution2D(32, 64, 5, pad: 2, name: "l2 Conv2D"),
new ReLU(name: "l2 ReLU"),
new MaxPooling(2, 2, name: "l2 MaxPooling"),
new Linear(7 * 7 * 64, 1024, name: "l3 Linear"),
new ReLU(name: "l3 ReLU"),
new Dropout(name: "l3 DropOut"),
new Linear(1024, 10, name: "l4 Linear")
);
```
- Samples:
・[**XOR**](https://github.com/harujoh/KelpNet/blob/master/KelpNet.Sample/Sample/Sample01.cs)
・[**CNN**](https://github.com/harujoh/KelpNet/blob/master/KelpNet.Sample/Sample/Sample06.cs)
・[**AlexNet**](https://github.com/harujoh/KelpNet/blob/master/KelpNet.Sample/Sample/Sample19.cs)
・[**VGG**](https://github.com/harujoh/KelpNet/blob/master/KelpNet.Sample/Sample/Sample15.cs)
・[**ResNet**](https://github.com/harujoh/KelpNet/blob/master/KelpNet.Sample/Sample/Sample17.cs)
・[**Others**](https://github.com/harujoh/KelpNet/tree/master/KelpNet.Sample)
- SampleData:
・MNIST
・FashionMNIST
・CIFAR 10/100
- Importable:
・CaffeModel
・ChainerModel
・ONNXModel
## Features
- Uses the same "Define by Run" approach as PyTorch and Keras.
- No libraries are used for matrix operations, so all algorithms are readable.
- OpenCL is used for parallel processing, so processing can be parallelized not only on GPUs, but also on CPUs, FPGAs, and various other computing devices.
> * Additional installation of the corresponding driver may be required to use OpenCL.
> - Intel CPU or GPU: https://software.intel.com/en-us/articles/opencl-drivers
> - AMD CPU or GPU: https://www.amd.com/en/support
> - Nvidia GPU: https://developer.nvidia.com/opencl
### Advantages of being built in C#.
- Easy to set up a development environment and easy to learn for beginners in programming.
- There are many options for visual representation of processing results, such as the .Net standard Form and Unity.
- Development for various platforms such as PCs, mobile devices, and embedded devices is possible.
## How to contact us
If you have any questions or concerns, even minor ones, please feel free to use Issue.
If you want to communicate with us easily, please contact us via X(Twitter).
You can also check the current development status on X(Twitter).
X(Twitter): https://twitter.com/harujoh
## System Requirements
Libraries: .NET Standard 2.0 or 2.1
Samples: .NET Framework 4.6.1
## Implemented Functions
- Connections:
・Convolution2D
・Deconvolution2D
・EmbedID
・Linear
・LSTM
- Activations:
・ELU
・LeakyReLU
・ReLU
・ReLU6
・Sigmoid
・Tanh
・Softmax
・Softplus
・Swish
・Mish
- Poolings:
・AveragePooling2D
・MaxPooling2D
- Normalize:
・BatchNormalization
・LRN
- Noise:
・Dropout
・StochasticDepth
- LossFunctions:
・MeanSquaredError
・SoftmaxCrossEntropy
- Optimizers:
・AdaBound
・AdaDelta
・AdaGrad
・Adam
・AdamW
・AMSBound
・AMSGrad
・MomentumSGD
・RMSprop
・SGD