earth object.
The plot shows model selection, cumulative distribution
of the residuals, residuals versus fitted values, and the residual QQ plot.## S3 method for class 'earth':
plot(x = stop("no 'x' arg"), which = 1:4,
info = FALSE, delever = FALSE, pearson = FALSE, level = 0, versus = NULL,
nresponse = 1, npoints = 1000, id.n = 3,
labels.id = rownames(residuals(object, warn=FALSE)),
center = TRUE, loess.f = .5,
do.par = length(which) > 1, xlim = NULL, ylim = NULL,
main = NULL, cex.main = 1.2, caption = if(do.par) NULL else "",
xlab = NULL, ylab = NULL,
pch = 20, col.line = "lightblue",
col.loess = NULL, lwd.loess = NULL, col.cv = "red",
col.qq = col.line, col.grid = "lightgray", col.points = 1, cex.points = NULL,
shade.pints = "mistyrose2", shade.cints = "mistyrose4",
cum.grid = "percentages",
col.rsq = NA, col.residuals = NA, nresiduals = NA,
# following are passed to plot.model.selection
legend.pos = NULL, cex.legend = NULL, col.grsq = 1, col.infold.rsq = 0,
col.mean.infold.rsq = 0, col.mean.oof.rsq = "palevioletred",
col.npreds = if(is.null(object$cv.oof.rsq.tab)) 1 else 0,
col.oof.labs = 0, col.oof.rsq = "mistyrose2", col.oof.vline = col.mean.oof.rsq,
col.pch.cv.rsq = 0, col.pch.max.oof.rsq = 0, col.sel.grid = 0,
col.vline = col.grsq, col.vseg = 0, lty.grsq = 1, lty.npreds = 2,
lty.rsq = 5, lty.vline = 3, col.legend = NA,
...)1:4.
1) Model selection (GRSq versus number of terms)
2) Cumulative distribution of abs residuals
3) Residuals versus fitted
4) QQ plot of residuals
5) Abs residuals versus fitted
6) Abs residuals versus log fiFALSE.
(1) Plot the distribution of the residuals along the bottom of the plot.
(2) Show the Spearman Rank Correlation of the absolute residuals
with the fitted valueFALSE.
Divide the residuals by sqrt(1 - h_ii),
where h_ii are the diagonal entries of the hat matrix.
The hat matrix here is from the linear fit on earth's basis matrix bx.
(NoFALSE.
Divide each residual by its estimated standard deviation.
Requires that the model was built with the varmod.method argument,
because we need the residual model to get the standard deviations.
Doe0 meaning do not plot the bands.
A typical value is level=.95.
Requires that the model was built with the varmod.method argument.
The color NULL Default. Plot the residuals versus the fitted values
(the log fitted values when which=6 to 8).
"" Residuals versus the predictors (multiple plots i-1 for all.
Default is 1000 (not all to reduce over-plotting).
A systematic sample of size nresiduals is taken but
the largest few id.n residuals will be labeled in the plot.
Default is 3.rownames(residuals(x)).
Only used if id.n > 0..5.
Lower values make the line bumpier.
This argument is passed as f in the internal call to lowesTRUE, start a new page and call par as appropriate.
Default is length(which) > 1, i.e., call par if
drawing more than one plot.
Use FALSE to use the cuwhich has more than one element,
this argument applies only to the Model Selection plot.
In the model selection plot,
the special value min=-1 means theNULL, meaning generate the headings automatically.1.2.
Used only if do.par is TRUE (default).if(do.par) NULL else "". One of:
"string" string
"" no caption
NULL generate a caption automatically.NULL, meaning label the x axis automatically.NULL, meaning label the y axis automatically.20."lightblue"
Use 0 for no RSq line.loess line.
Default is NULL, meaning automatic."red".
Color of cross validation line in the residuals plot.
This is the residual of the mean out-fold-predicted value.col.line.
Use 0 for no QQ line."lightgray". Use 0 for no grid.1NULL, meaning automatic.level argument was used.
Color of the prediction intervals.
Default is "mistyrose2"level argument was used.
Color of the confidence intervals.
Default is "mistyrose4""none" no grid on Cumulative Distribution plot
"grid" add grid
"percentages" (default) add grid and percentage labels to quantile lines.col.line instead.
Default is NA.col.points instead.
Default is NA.npoints instead.
Default is NA.
The following are passed to plot.model.selection.NULL, meaning automatic.
Else specify c(x,y) in user coordinates,
or use "topleft" etc. as explained in legend.NULL, meaning automatic.1.
Use 0 for no GRSq line.nfold and keepxy were used
in the original call to earth.
Default is 0, lines not plotted.0, line not plotted.
Applies only if nfold and keepxy were used
in the original call to earth.nfold and keepxy were used
in the original call to earth.
Default is "palevioletred".
Uoof.rsq's are displayed.
Use 0 for no "number of predictors" plot.oof.rsq lines.
Default is 0, no labels.nfold and keepxy were used
in the original call to earth.
Default is "mistyrose2", a pale pink.
Uoof.rsq in the Model Selection plot.
Default is col.mean.oof.rsq.oof.rsq line to indicate the cv.rsq.
for that fold
(i.e., it is plotted at the number of terms selected by the in-fold GCV).
Default is 0, point not plotted.oof.rsq line to indicate the
maximum oof.rsq for that fold.
Default is 0, point not plotted.0, no grid.
Try something like "lightgray", "linen", or "seashell".
See also col.cum.grid, for the grid in the Cumulacol.grsq.
This will be at the maximum GRSq unless pmethod="none".
Use 0 for no vertical line.0.
Color of triangular marker at top of vertical line for best GRSq.12.5.3.1.
Use 0 for no legend.earth,
plot.earth.models,
plotd,
plotmodata(ozone1)
earth.mod <- earth(O3 ~ ., data = ozone1, degree = 2)
plot(earth.mod)Run the code above in your browser using DataLab