Neural Network Prototype and Iris dataset example

Platform	Build status
Linux OSX

Building and testing

mkdir build && cd build && cmake .. && cmake --build .

libnnp

libnnp implements a simple feedforward neural network.

Network layers can be formed by making a specialization of the nnp::ComputationalLayer class template. Multiple layers can be appended with the nnp::TupleNetwork class template. Adding a loss layer to a nnp::TupleNetwork and calling the propagate() function with the appropriate parameters trains the network a single iteration. propagate() also has an overload to check the loss without back propagation to use with a validation set. Calling the forward() function of nnp::TupleNetwork returns the output tensor from the outermost layer. This can be used at test time.

Iris dataset example

After the project is built, run the program by passing it the path of the iris dataset.

./build/example/iris/iris_training example/iris/iris.data