plot.validann: Plot ANN validation results.
In validann: Validation Tools for Artificial Neural Networks

Description Usage Arguments Details See Also Examples

Plot method for objects of class ‘validann’. Produces a series of plots used for validating and assessing ANN models based on results returned by validann.

1
2
3

## S3 method for class 'validann'
plot(x, obs, sim, gof = TRUE, resid = TRUE, sa = TRUE,
  display = c("multi", "single"), profile = c("all", "median"), ...)

`x`	object of class ‘validann’ as returned by `validann`. This is a list comprising metrics and statistics that can be used for validating ANN models.
`obs, sim`	vectors comprising observed (`obs`) and simulated (`sim`) examples of a single response variable used for computing `x` object.
`gof`	logical; should goodness-of-fit plots be produced? Default = TRUE.
`resid`	logical; should residual analysis plots be produced? Default = TRUE.
`sa`	logical; should input sensitivity analysis plots be produced? Default = TRUE.
`display`	character string defining how plots should be displayed. The default is “multi” where multiple plots are displayed together according to whether they are goodness-of-fit, residual analysis or sensitivity analysis plots. For “single”, each plot is displayed on its own. If the session is interactive, the user will be asked to confirm a new page whether `display` is “single” or “multi”.
`profile`	character string defining which structural validity Profile method outputs should be plotted. The default is “all” where outputs corresponding to 5 summary statistics are plotted together with the median predicted response for each input value. For “median”, only the median response is plotted.
`...`	Arguments to be passed to plot (not currently used).

This function can be invoked by calling plot(x, obs, sim) for an object x of class ‘validann’.

To produce plots for all types of validation metrics and statistics, gof, resid and sa must be TRUE and corresponding results must have been successfully computed by validann and returned in object x.

If gof is TRUE, a scatter plot, Q-Q plot and time/sample plot of observed (obs) versus predicted (sim) data are produced.

If resid is TRUE and x$residuals is not NULL, plots of the model residuals are produced including histogram, Q-Q plot (standardized residuals compared to standard normal), autocorrelation (acf), partial autocorrelation (pacf), standardized residual versus predicted output (i.e. sim) and standardized residual versus time/order of the data.

If sa is TRUE and x$y_hat is not NULL, model response values resulting from the Profile sensitivity analysis are plotted against percentiles of each input. If x$rs is not NULL, the relative sensitivities of each input, as computed by the partial derivative (PaD) sensitivity analysis, are plotted against predicted output.

Setting gof, resid and/or sa to FALSE will ‘turn off’ the respective validation plots.

validann

## Build ANN model and compute replicative and structural validation results
data("ar9")
samp <- sample(1:1000, 200)
y <- ar9[samp, ncol(ar9)]
x <- ar9[samp, -ncol(ar9)]
x <- x[, c(1,4,9)]
fit <- ann(x, y, size = 1, act_hid = "tanh", act_out = "linear", rang = 0.1)
results <- validann(fit, x = x)
obs <- observed(fit)
sim <- fitted(fit)

## Plot replicative and structural validation results to the current device
## - a single page for each type of validation
plot(results, obs, sim)

## Plot results to the current device - a single page for each plot
plot(results, obs, sim, display = "single")

## Plot replicative and structural validation results to single file
pdf("RepStructValidationPlots.pdf")
plot(results, obs, sim)
dev.off()

## Get predictive validation results for above model based on a new sample
## of ar9 data.
samp <- sample(1:1000, 200)
y <- ar9[samp, ncol(ar9)]
x <- ar9[samp, -ncol(ar9)]
x <- x[, c(1,4,9)]
obs <- y
sim <- predict(fit, newdata = x)
results <- validann(fit, obs = obs, sim = sim, x = x)

## Plot predictive results only to file
pdf("PredValidationPlots.pdf")
plot(results, obs, sim, resid = FALSE, sa = FALSE)
dev.off()