Machine learning library and tool with terminal user interface written in rust. It supports OpenCL layers and CPU layers. Core math matrix library is ndarray which use matrixmultiply crate for CPU matrix multiplication.
- Fast optimized computations
- User-friendly API
- Provide utility terminal application to create, train, manage models with user-friendly terminal interface
OpenCL support is based on ocl crate. It is optional feature and enabled by default.
- FullyConnected layer
- Euclidean Loss, Softmax Loss
- Optimizers: Adam, RMSProp, AdaGrad, AdaDelta
- Async parallel data loading
- Protobuf, CSV dataloaders
- (De)Serializing neural network state to protobuf
- (De)Serializing neural network configuration net yaml file
- Activation functions : sigmoid, tanh, relu, leaky_relu
XOR function approximation
cargo run --release --example xor
XOR function approximation with OpenCL support
cargo run --release --example xor_ocl
For MNIST digits classification example read MNIST example readme
- Conv2D layer
- RNN + LSTM
- Residual block
- OpenCL optimization
Apache License Version 2.0