Plot sensitivities of a neural network model

Function to plot the sensitivities created by HessianMLP.

Usage

SensMatPlot(
  hess,
  sens = NULL,
  output = 1,
  metric = c("mean", "std", "meanSensSQ"),
  senstype = c("matrix", "interactions"),
  ...
)

Arguments

hess: HessMLP object created by HessianMLP.
sens: SensMLP object created by SensAnalysisMLP.
output: numeric or character specifying the output neuron or output name to be plotted. By default is the first output (output = 1).
metric: character specifying the metric to be plotted. It can be "mean", "std" or "meanSensSQ".
senstype: character specifying the type of plot to be plotted. It can be "matrix" or "interactions". If type = "matrix", only the second derivatives are plotted. If type = "interactions" the main diagonal are the first derivatives respect each input variable.
...: further argument passed similar to ggcorrplot arguments.

Value

a list of ggplots, one for each output neuron.

Details

Most of the code of this function is based on ggcorrplot() function from package ggcorrplot. However, due to the inhability of changing the limits of the color scale, it keeps giving a warning if that function is used and the color scale overwritten.

Examples

## Load data -------------------------------------------------------------------
data("DAILY_DEMAND_TR")
fdata <- DAILY_DEMAND_TR
## Parameters of the NNET ------------------------------------------------------
hidden_neurons <- 5
iters <- 100
decay <- 0.1

################################################################################
#########################  REGRESSION NNET #####################################
################################################################################
## Regression dataframe --------------------------------------------------------
# Scale the data
fdata.Reg.tr <- fdata[,2:ncol(fdata)]
fdata.Reg.tr[,3] <- fdata.Reg.tr[,3]/10
fdata.Reg.tr[,1] <- fdata.Reg.tr[,1]/1000

# Normalize the data for some models
preProc <- caret::preProcess(fdata.Reg.tr, method = c("center","scale"))
nntrData <- predict(preProc, fdata.Reg.tr)

#' ## TRAIN nnet NNET --------------------------------------------------------
# Create a formula to train NNET
form <- paste(names(fdata.Reg.tr)[2:ncol(fdata.Reg.tr)], collapse = " + ")
form <- formula(paste(names(fdata.Reg.tr)[1], form, sep = " ~ "))

set.seed(150)
nnetmod <- nnet::nnet(form,
                      data = nntrData,
                      linear.output = TRUE,
                      size = hidden_neurons,
                      decay = decay,
                      maxit = iters)
#> # weights:  21
#> initial  value 2487.870002 
#> iter  10 value 1587.516208
#> iter  20 value 1349.706741
#> iter  30 value 1333.940734
#> iter  40 value 1329.097060
#> iter  50 value 1326.518168
#> iter  60 value 1323.148574
#> iter  70 value 1322.378769
#> iter  80 value 1322.018091
#> final  value 1321.996301 
#> converged
# Try HessianMLP
H <- NeuralSens::HessianMLP(nnetmod, trData = nntrData, plot = FALSE)
NeuralSens::SensMatPlot(H)

S <- NeuralSens::SensAnalysisMLP(nnetmod, trData = nntrData, plot = FALSE)
NeuralSens::SensMatPlot(H, S, senstype = "interactions")