Estimate a Neural Network
To estimate a model select the type (i.e., Classification or Regression), response variable, and one or more explanatory variables. Press the
Estimate button or
CMD-enter on mac) to generate results. The model can be “tuned” by changing the
Size (i.e., the number of nodes in the hidden layer) and by adjusting the
Decay rate. The higher the value set for
Decay, the higher the penalty on the size of (the sum of squares of) the weights. When
Decay is set to 0, the model has the most flexibility to fit the (training) data accurately. However, without
Decay the model is also more likely to overfit.
The best way to determine the optimal values for
Decays is to use Cross-Validation. In radiant, you can use the
cv.nn function for this purpose. See the documentation for more information.
Add code to Report > Rmd to (re)create the analysis by clicking the icon on the bottom left of your screen or by pressing
ALT-enter on your keyboard.
If either a
Garson plot was created it can be customized using
ggplot2 commands (e.g.,
plot(result, plots = "garson", custom = TRUE) + labs(title = "Garson plot")). See Data > Visualize for details.
To add, for example, a title to a network plot use
title(main = "Network plot"). See the R graphics documentation for additional information.
For an overview of related R-functions used by Radiant to estimate a neural network model see Model > Neural network.
The key function from the
nnet package used in the
nn tool is