SPARTAAS (version 1.2.4)

mapclust: Divise hierarchical Clustering using Spatialpatches algorithm.

Description

This function performs a divisive hierarchical clustering on a regionalised variable using the spatial patches algorithm (Woillez et al. 2007; Woillez, Rivoirard and Petitgas 2009). It is a top-down hierarchical clustering with a geographical constraint. It is possible to cut the tree by clicking on the dendrogram at the desired level. The results include a description of the clusters and graphics. When slicing the dendrogram, you can look at the two plots (WSSPlot and AveSilPlot) that show the relatively good quality of the partitions. The first shows the Within Sum of Square (WSS) for each partition and you can use the elbow approach to select a partition. The second graph shows the average silhouette width. This index is between -1 and 1. The closer it is to 1, the better the partition. See silhouette.

If you want to cut the dendogram to a different dlim or number of clusters, you can do so without re-running mapclust() with mapclust_cut_tree.

Usage

mapclust(
    coord,
    var,
    label = "Nolabel",
    iter = 20,
    Plabel = TRUE,
    lonlat = TRUE,
    positive_var = FALSE,
    n = NULL
)

Value

the function returns a list.

Plot:

dendrogram

The global dendrogram (hclust object)

dendrogram_ggplot

The global dendrogram (ggplot2 object)

cuttree

The cut dendrogram

map

The map of the selected partition

AveSilPlot

The average silhouette width plot (for each partition)

WSSPlot

The Within Sum of Square plot (for each partition)

silhouette

The silhouette plot of the selected partition

Value:

X

The x-coordinate data you have used

Y

The y-coordinate data you have used

var

The regionalized variable(s) data you have used

label

The label vector you have used

density

The estimated density based on var. Equal to var if you used a unidimensional density variable.

cluster

The cluster vector of the selected partition

Plabel

Logical parameter to activate or not the printing of labels on the dendrogram

fullhist

The compositional cluster for each observation

hist

The compositional cluster without duplicates (corresponds to the split on the dendogram)

dlim

The vector of the different limit distances

cutdlim

The select dlim for the cut of the current partition

DiMatrix

The weighted euclidean distance matrix

silhouetteData

The silhouette data of the selected partition

AveSilData

The average silhouette value for each partition

Moran

The Moran index for each group for each partition

lonlat

Logical parameter, whether your coordinates are in latitude-longitude format or not

Arguments

coord

The x (longitude) and y (latitude) coordinates of the data.frame or matrix dimension 2.

var

The regionalised variable(s) of interest

label

(optional) The names of the samples or an identifier. Must be a factor.

iter

The number of iterations. The number of different dlim you want to test (must be greater than 10).

Plabel

Logical parameter to activate or not the printing of labels on the dendrogram.

lonlat

Logical parameter to activate or not the cartography in lonlat system with leaflet (base map).

positive_var

logical parameter that defines whether your variable of interest is positive or not.

n

Number of groups. If NULL, you can select the number of groups by clicking on the dendrogram.

Author

A. COULON L. BELLANGER P. HUSI

Details

Dlim is the selected minimum distance from the sample to the patch centre: to identify patches (units are those of coordinates). The dlim is automatically initialised and does not need to be set by the user. The minimum data is a data frame or matrix with at least 3 columns.

References

Bellanger L., Coulon A. and Husi P. (2021) Determination of cultural areas based on medieval pottery using an original divisive hierarchical clustering method with geographical constraint (MapClust), Journal of Archaeological Science, Volume 132 tools:::Rd_expr_doi("10.1016/j.jas.2021.105431").

Bellanger L., Husi P., Laghzali Y. (2015). Spatial statistic analysis of dating using pottery: an aid to the characterization of cultural areas in West Central France. In : Traviglia A. ed., Across Space and Time, Proceedings of the 41th International Conference on Computer Applications and Quantitative Methods in Archaeology (CAA-2013), Perth (Australie), Amsterdam University Press : 276-282.

Woillez M., Poulard J.C., Rivoirard J., Petitgas P., Bez N. (2007). Indices for capturing spatial patterns and their evolution in time, with application to European hake (Merluccius merluccius) in the Bay of Biscay. ICES J. Mar. Sci. 64, 537-550.

Woillez M., Rivoirard J. and Petitgas P. (2009) Notes on survey-based spatial indicators for monitoring fish populations, Aquatic Living Resources, 22 :155-164.

Examples

# \donttest{
###################################
## loading data
 library(SPARTAAS)
 data(datarcheo)
 data(datacancer)

###################################
### Example: 1
## Function "mapclust"
# object <- mapclust( coord = ..., var = ..., label = ...)

classification <- mapclust(datarcheo$coord, datarcheo$var, datarcheo$label, n=4)

#Global dendrogram
 classification$dendrogram
#Cut dendrogram
 classification$cuttree
#silhouette of selected partition
 classification$silhouette


#You can cut the dendrogram for another dlim
 NewCut <- mapclust_cut_tree(classification, dlim=0.30)

#See evaluation using Silhouette width by running:
 NewCut$silhouette
 #If the plot is empty try to increase the height of the window (full screen)

#See summary of the data by running:
 summary(NewCut$silhouetteData)


###################################
## kmeans comparison
# pepare data (only geographical data)
 datakmeans <- datarcheo$coord

#kmeans
 number_cluster <- 4
 cl <- kmeans(datakmeans, number_cluster)
 plot(datakmeans, col = cl$cluster)


# }