Vectorflow is a minimalist neural network library optimized for sparse data and single machine environments.
Original blog post here.
The library is distributed as a dub package. Add vectorflow to the dependencies section of your dub.json:
"vectorflow": "~>1.0.0"
The library itself doesn't have any dependencies. All you need is a recent D compiler.
LDC is the recommended compiler for the fastest runtime speed.
Tested on:
- Linux, OSX
- LDC version: >= 1.1.1
- DMD version: >= 2.073.1
If you're new to D, keep reading. You will need dub (the D package manager) and LDC (the LLVM-based D compiler).
brew install dub
brew install ldc
dub can be downloaded here (or follow the instructions on this page).
LDC can be installed by running:
snap install --classic --edge ldc2
To run the RCV1 example (sparse logistic regression):
cd examples && ./compile_run.sh rcv1.d
To run the tests:
dub test
vectorflow uses ddoc.
One way of building and serving the documentation locally (you will need libevent for serving) is:
dub build -b ddox && dub run -b ddox
Or use your favorite DDOC compiler.
Please also refer to the repo wiki.