OptimStride: Compute a Series of Partitions/Clusterings
In vegsoup: Classes and Methods for Phytosociology

Description Usage Arguments Details Value Methods Author(s) References See Also Examples

Compute a series of clusterings (strides) by different methods and quantify their quality using a Fisher test.

OptimStride(x, k, ft.threshold = 1e-3, alternative = "greater",
            method = c("ward", "flexible", "pam", "kmeans", "wards",
                       "fanny", "FCM", "KM", "optpart"), fast = FALSE, ...)

## S4 method for signature 'VegsoupOptimstride'
method(x, ...)

## S4 method for signature 'VegsoupOptimstride'
stride(x, method, ...)

## S4 method for signature 'VegsoupOptimstride'
threshold(x, ...)

## S4 method for signature 'VegsoupOptimstride'
optimclass1(x, ...)

## S4 method for signature 'VegsoupOptimstride'
optimclass2(x, threshold = 2, ...)

## S4 method for signature 'VegsoupOptimstride'
getK(x)

## S4 method for signature 'VegsoupOptimstride'
peaks(x, ...)

## S4 method for signature 'VegsoupOptimstride'
which.max(x)

`x`	for `OptimStride` a `VegsoupPartition` object, otherwise a `VegsoupOptimstride` object
`k`	maximum number of cluster to compute (`seq(1:k)`)
`ft.threshold`	threshold value of Fisher test, defaults to a not very strict value of `1e-03`.
`alternative`	alternative indicates the alternative hypothesis of the Fisher exact test and must be one of "two.sided", "greater" or "less" (see `FisherTest` and fisher.test).
`threshold`	threshold value for OptimClass2. See ‘Details’
`method`	any method supported by `VegsoupPartition`. For `OptimStride` if method is missing, "ward", "flexible", "pam", "kmeans", "wards", "fanny", "FCM", and "KM" are computed. Computationally expensive methods, such as "isopam" and "optpart" are omitted to reduce run duration.
`fast`	accelerate computations using package `parallel`. If `FALSE` a progress bar is printed. Depending on the dimension of the community matrix and the value of k, computations may consume a serious amount of time. If the host system provides 4 or more cores, setting`fast = TRUE` will significantly reduce run time.
`...`	additional arguments passed to `VegsoupPartition`, otherwise not used.

The implementation follows the method of Tichy et al. (2010) and uses a Fisher test (FisherTest) to identify the number of ‘faithful’ species. This sensitivity of this statistic can be controlled by setting a threshold value. Per default this value (argument ft.threshold) is set to it's lowest meaningful bound 1e-3.

Partitioning/clustering methods, type of distance matrix and standardization are taken from the input object. Generic methods for 'VegsoupOptimstride' are detailed in the ‘Methods’ section.

`Optimstride`	`VegsoupOptimstride` object
`method`	character. The computed methods.
`stride`	list. A list with length equal to `length(method(x))`. Each nested list gives the number of significant indicator species (faithful species) for each partition for a particular value of `k`.
`optimclass1, optimclass2`	matrix with dimnames. Rows are methods and columns are the respective values of `k`.
`peaks`	peaks of the curve.

method: signature(obj = "VegsoupOptimstride") : retrieves the names of the computed method(s). These are the method arguments as accepted by VegsoupPartition.
stride: signature(obj = "VegsoupOptimstride") : returns the number of faithful species for each partition and number of k.
threshold: signature(obj = "VegsoupOptimstride") : retrieves the threshold values set for the Fisher test.
optimclass1, optimclass2: signature(obj = "VegsoupOptimstride") : returns a matrix with the counts of faithful species, those achieving ft.trshold.
getK: signature(obj = "VegsoupOptimstride") : the computed maximum value of k, the length of each stride.
peaks: signature(obj = "VegsoupOptimstride") : the peaks along the stride.
which.max: signature(obj = "VegsoupOptimstride") : returns the number of k that coincides with the highest number of faithful species, per method.

Roland Kaiser

Tichy, L., Chytry, M., Hajek, M., Talbot, S.S., and Botta-Dukat, Z. (2010). Optimclass: Using species-to-cluster fidelity to determine the optimal partition in classification of ecological communities. Journal of Vegetation Science, 21(2):287-299.

VegsoupPartition, fidelity

require(vegsoup)

# a dummy example using simulated data
x <- coenoflex(50, 100)

x <- OptimStride(x, k = 10, method = c("ward", "flexible", "pam"))

summary(x)

# get the computed clustering methods
method(x)

# the threshold of the Fisher test
threshold(x)

# matrix of results for OptimClass1
optimclass1(x)
boxplot(t(optimclass1(x)))

# the number of faithful species for each partition and method
# warning, method dispatch will break if package optpart is loaded
stride(x)
stride(x, method = "flexible") # for method flexible

# plot method for class VegsoupOptimstride
plot(x)
plot(x, method = "flexible")

# get k where the curve peaks
peaks(x)

# get k with the maximum number of faithful species
which.max(x)