test.parameters {EMMAgeo}

Evaluate influence of model parameters.

Description

All possible combinations of number of end-members and weight transformation limits are used to perform EMMA and evaluate the absolute and relative measures of individual model performance.

Usage

test.parameters(
  X,
  q,
  l = 0,
  c = 100,
  rotation = "Varimax",
  plot = FALSE,
  legend,
  multicore = FALSE,
  ...
)

Arguments

X

Numeric matrix, input data set with m samples (rows) and n variables (columns).

q

Numeric vector, numbers of end-members to be modelled, e.g., 2:10.

l

Numeric vector specifying the weight tranformation limit, i.e. quantile; default is 0.

c

Numeric scalar specifying the constant sum scaling parameter, e.g., 1, 100, 1000; default is 0.

rotation

Character scalar, rotation type, default is "Varimax".

plot

Character scalar, optional graphical output of the results as keyword (see details). All plots except "ol" are colour-coded bitmaps of q, l and the specified test parameter and line-plots the specified parameter vs. q.

legend

Character scalar, specifying legend position (cf. legend). If omitted, no legend will be plotted, default is no legend.

multicore

Logical scalar, optionally ditribute calculations to all available cores of the computer, default is TRUE.

...

Additional arguments passed to the plot function (see details).

Details

The mean total explained variance mRt may be used to define a maximum number of meaningful end-members for subsequent modelling, e.g. as the number of end-members, which reaches the first local mRt maximum.
Overlapping is defined as one end-member having its mode within the "area" of any other end-member, which is genetically not explainable.
Keywords to specify, which tested parameter will be plotted: "mEm" (mean absolute row-wise error), "mEn" (mean absolute column-wise error), "mRm" (mean relative row-wise error), "mRn" (mean relative column-wise error), "mRt" (mean relative total error) and "ol" (number of overlapping end-members).
Since the function returns two plots (except for option "ol"), additional graphical parameters must be specified as vector with the first element for the first plot and the second element for the second plot. If graphical parameters are natively vectors (e.g. a sequence of colours), they must be specified as matrices with each vector as a row. A legend can only be added to the second plot. Colours only apply to the second plot as well. If colours are specified, colour should be used instead of col. See example section for further advice.

Value

List with result objects

mEm

Absolute row-wise model error.

mEn

Absolute column-wise model error.

mRm

Mean row-wise explained variance.

mRn

Mean column-wise explained variance.

mRt

Mean total explained variance.

ol

Number of overlapping end-member loadings.

q.max

Maximum number of meaningful end-members.

Author(s)

Michael Dietze, Elisabeth Dietze

References

Dietze E, Hartmann K, Diekmann B, IJmker J, Lehmkuhl F, Opitz S, Stauch G, Wuennemann B, Borchers A. 2012. An end-member algorithm for deciphering modern detrital processes from lake sediments of Lake Donggi Cona, NE Tibetan Plateau, China. Sedimentary Geology 243-244: 169-180.

See Also

EMMA

Examples

## load example data set
data(example_X)

## truncate the data set for faster computation
X.trunc <- X[1:20,]

## define test parameters
q <- 2:8 # number of end-members
l <- seq(from = 0, to = 0.3, by = 0.1)

## test parameter influence and plot mean total explained variance
TP <- test.parameters(X = X.trunc, q = q, l = l, plot = "mRt",
                      legend = "bottomright", cex = 0.7,
                      multicore = FALSE,
                      colour = rgb((1:7) / 7, 0.9, 0.2, 1))

## show maximum number of end-members
TP$q.max