neural-network-approximator

Command line application to generate neural networks as function approximators from sets of data points.

This is an application created to experiment with the OpenNN library for building neural networks.

Requirements

• C++20 compiler.
• CMake 3.22.0.
• Boost::program_options 1.74.0.

Usage

Usage: neural-network-approximator --config [PATH] --data-set [PATH] --output [PATH]:
  --help                show help.
  --config arg          set configuration file path.
                        Available parameters:
                        * learning-rate
                        * loss-goal
                        * max-epoch
                        * regularization=[None L1 L2]
                        * regularization-rate
                        * layers=[NUMBER_OF_NEURONS] 
                        * layers=[ACTIVATION_FUNCTION=[relu, tanh, step, 
                        sigmoid, linear]]
                        * layers=...
  --data-set arg        set data set file path.
  --output arg          set output files path.

Examples

cos(x) with 1 hidden layer = [6 tanh]:

x² with 1 hidden layer = [10 relu]:

√x with 1 hidden layer = [10 relu]: