neural-net

Create a working prototype of a neural network using schotastic gradient descent. Designed for the networks to be easily customisable and exportable.

Includes custom, light-weight linear algebra library.

How to run

mkdir build
cd build
cmake ..
make
./main_app

Improvements yet to be completed:

• add documentation
• add test suite / more examples of simple tests
• optimise efficiency (remove obsolete variables)
• separate into mini batches
• more descriptive error messages to stderr
• remove unecessary imports in certain fiels
• make class variables private as necessary (and convert to setters/getters)
• refactor importing neural network (currently includes repeated code)
• more comprehensive documentation
• make (input, target) a tuple / std::pair
• use strategy / builder / template pattern to allow for different activation functions, cost functions, etc.
• create a data loader class / handles all the feeding data into network and getting predictions
• linear_algebra
  • make it easier to initalise vectors and matrices...
  • use templates (T extends number)