Use WebAssembly for Training and Inference
rainij opened this issue · comments
Currently everything is done in JavaScript. It might be good for performance if training and possibly inference too would be re-implemented in WASM.
Things to consider:
- Do some Bench-marking to not accidentally get slower.
- WASM implies that consumers of this library have to do something (e.g. adjust config of bundler). Give an example of how to use this Library with webpack.
- Implementation of the training and inference algorithms could probably be done in C or C++ and then translated to WASM by emscripten. The code for the algorithms shouldn't contain many lines or be complicated. So the code can be kept within this repository.