NN-TSV, NCEP neural network training and validation system
Vladimir M. Krasnopolsky · NOAA Institutional Repository · 2014
This Note describes neural network (NN) training and validation system (TVS) or NN-TVS developed at NCEP (EMC). Section 1 of this Note is an introductory section. Section 2 presents a brief NN tutorial containing limited discussion of basic ideas of NN technique and NN features implemented in the NN-TVS system. Some of the NN features discussed in Section 2 have not yet been implemented in the NN-TVS system and are expected to be implemented in following versions of the system. Section 3 contains a description of the NN-TVS structure and functional relationships between blocks. It also describes scripts necessary for running the system on NCEP computers. Section 4 contains a description of FORTRAN module neural.f90 that can be used to run the trained NN in applications. Section 5 contains appendixes with information about the data files used in NN-TVS (Appendix 1); also the functions and subroutines constituting the NN-TVS system are listed in Appendix 2. The text of the FORTRAN module neural.f90 is presented in Appendix 3. Appendix 4 explains the evaluation statistics that are used in the NN-TVS evaluations of NN training results.