-
Notifications
You must be signed in to change notification settings - Fork 19
0. Getting Started
The easiest way to get started with Bright Wire is to install it via Nuget.
Install-Package BrightWire.Net4
If you have a supported NVIDIA GPU (Kepler or better) and have installed CUDA Toolkit 8 on your machine you can add CUDA support with:
Install-Package BrightWire.CUDA.Net4.x64
Note: When using CUDA, make sure that the /LinearAlgebra/cuda/kernel.ptx file's is copied to the output directory (Properties/Copy To Output Directory).
The goal of the library is to provide an easy way to train and execute common machine learning algorithms from within the .NET ecosystem. Also, any machine learning algorithm that uses linear algebra should be executable on the GPU.
Tutorials and a guide to getting started are available here.
When using machine learning on small data sets the advantages of GPU based training is small to negative. There are considerable costs involved when moving data around between the CPU and the GPU and if those costs outweigh the cost of computation it is better to keep the computation on the CPU.
However the current trend is to use larger data sets with ever increasing parameters and it is in the large parameter setting that GPU based computation really shines, cutting the training and execution time significantly (the exact improvement is largely dependent on the GPU hardware).