This package provides neural network modules supporting sparse data in various levels using Torch.
- git clone https://github.com/ameenetemady/SparseNN/
- cd SparseNN
- luarocks make rocks/sparsenn-scm-1.rockspec
When it comes to neural networks, the support for sparsity is often limitted to input layers only (e.g. SparseLinear). This repository is intended for development of NN modules supporting sparsity in various layers (e.g. modules that preserve sparsity).
Depending on the problem structure a different "sparsity-aware" implementation maybe desired.
The first such implementation focused here is SparseBlock.