Deep Learning Library for JavaScript
Documents are under preparation. Currently, example of training MNIST digit recognition for node.js environment is provided. Code for distributed computing using web browser clients will be uploaded in near future.
github.io page Sukiyaki2 MILJS
Since this project is written in TypeScript, transpiling to JavaScript is necessary.
git clone https://github.com/mil-tokyo/sukiyaki2
cd sukiyaki2
npm install
npm run build
On npm install
, it installs node-opencl package for GPU computing which allows dramatically faster computation.
In my environment (Ubuntu 14.04 + NVIDIA CUDA 7.5), installation with node-opencl requires additional environment variables.
CPLUS_INCLUDE_PATH=/usr/local/cuda/include LIBRARY_PATH=/usr/local/cuda/lib64 npm install
Run on root directory
./example/mnist/prepare.py
./example/mnist/train.sh
By default, GPU is not used. To try fast computation,
./example/mnist/train.sh cl
They are in docs/example
directory.
Access to online demo from https://mil-tokyo.github.io/sukiyaki2/