v1.0.3

A change was made to the garson function to use more explicit calls to methods in the appendix of Goh 1995. The previous versions used matrix multiplication to add the contributions of the network connections, which is similar but not identical to Garson's approach.

v1.0.2

The functions were modified so they won't break if using neural networks trained with a skip layer. However, plotting and variable importance values may be unreliable. See the response in the issues section.

v.1.0.1

*fixed bug with only two variables in lekprofile

*garsons algorithm now works with one variable

*lekprofile stops with one variable

CRAN version

This is the version submitted to CRAN including functions for plotting, variable importance, sensitivity, and weights.