Clustering thermal histories models

Groups t-T paths into "path families" based on the Hausdorff or Fréchet distance between paths.

Usage

cluster_paths(
  x,
  k,
  dist = c("Hausdorff", "Frechet"),
  method = c("hclust", "kmeans", "pam", "dbscan", "hdbscan", "specc", "diana", "agnes",
    "clara", "fanny"),
  naming = c("asis", "GOF", "size"),
  warn = TRUE,
  threshold = 0.01,
  ...
)

Arguments

x

either an object of class "HeFTy" (output of read_hefty()), an object of "tTdiss" (output of path_diss()), or a data.frame containing the time, temperature, and segment columns of the modeled paths.

k

an integer scalar or vector with the desired number of groups. Ignored when dist equal to dbscan or hdbscan (see Note).

dist

character. Algorithm to calculate a dissimilarity matrix (distance) for lines; either Hausdorff (the default), or Frechet.

method

character. Clustering method to be applied. Currently implemented are

"hclust"

Hierarchical Clustering using stats::hclust(), the default)

"kmeans"

K-Means Clustering using stats::kmeans())

"pam"

Partitioning Around Medoids using cluster::pam()

"dbscan"

Density-based Spatial Clustering of Applications with Noise (DBSCAN) using dbscan::dbscan() (see Note)

"hdbscan"

Hierarchical DBSCAN using dbscan::hdbscan() (see Note)

"specc"

Spectral Clustering using kernlab::specc()

"agnes"

Agglomerative hierarchical clustering using cluster::agnes()

"diana"

Divisive hierarchical clustering using cluster::diana()

"clara"

Clustering Large Applications using cluster::clara()

"fanny"

Fuzzy Analysis Clustering using cluster::fanny()

naming

character. Naming scheme for clusters. One of "asis" (output of underlying cluster algorithm), "GOF" (ranks of the mean GOF values within clusters), and "size" (ranks of the size of clusters). Outliers detected by dbscan of hdbscan will always named as 0.

warn

logical. Should there be a warning message if at least one cluster contains less than (threshold * 100)% of the total paths?

threshold

numeric. The significance threshold as a fraction of the total amount of paths (0.01 by default). If one cluster contains less paths per total paths than this value, a warning message is created. Ignored if warn is FALSE.

...

additional arguments passed to cluster method.

Value

a data.frame with the path segment (integer) and cluster (factor)

Details

If you want to use a different clustering method that is not built in the current version of thermoclustr, you can use the distance matrix produced by path_diss() and feed your cluster algorithm.

Note

that dbscan and hdbscan methods require eps and minPts arguments. Optimal eps values can be visually estimated from the "knee" in a k-nearest neighbor distance plot using dbscan::kNNdistplot()'.

See also

path_diss() for calculating dissimilarities and path_nbclust() for determining the optimal number of clusters.

Examples

data(tT_paths)
tT_paths$paths <- subset(tT_paths$paths, Comp_GOF >= 0.4)

# cluster the paths
cluster_paths(tT_paths, k = 3)
#>    segment cluster
#> 1        1       1
#> 2       10       2
#> 3      104       2
#> 4       11       3
#> 5       12       2
#> 6      122       3
#> 7       13       3
#> 8       14       2
#> 9       15       1
#> 10     152       1
#> 11      16       2
#> 12     164       3
#> 13     166       3
#> 14      17       2
#> 15     172       1
#> 16     177       1
#> 17      18       2
#> 18      19       3
#> 19     195       3
#> 20       2       2
#> 21      20       2
#> 22      21       1
#> 23      22       2
#> 24      23       1
#> 25      24       2
#> 26      25       3
#> 27     259       2
#> 28      26       3
#> 29     268       3
#> 30      27       3
#> 31     274       3
#> 32     276       2
#> 33      28       1
#> 34      29       1
#> 35     290       3
#> 36     291       3
#> 37     292       3
#> 38     293       2
#> 39       3       2
#> 40      30       2
#> 41      31       2
#> 42      32       2
#> 43      33       3
#> 44      34       3
#> 45      35       1
#> 46      36       1
#> 47      37       3
#> 48      38       1
#> 49      39       2
#> 50       4       3
#> 51      40       3
#> 52      42       3
#> 53      43       3
#> 54      44       1
#> 55      45       1
#> 56      46       3
#> 57      47       3
#> 58      48       1
#> 59      49       3
#> 60       5       2
#> 61      50       3
#> 62      51       3
#> 63      55       3
#> 64       6       2
#> 65      63       2
#> 66       7       2
#> 67       8       2
#> 68       9       1