Skip to contents

Distance matrix from raw variables

Source: R/clustering.R
tda_pdatd.Rd

pdatd: builds a case-by-case distance matrix directly from a data matrix, the way dist does – Euclidean or city-block distance, the count of variables on which two cases differ, or a dissimilarity index. The result feeds directly into tda_cluster or tda_mds. Verified against dist for Euclidean and city-block (Manhattan) distance: exact match.

Usage

tda_pdatd(
  data,
  measure = 1,
  variables = NULL,
  weights = NULL,
  options = list(),
  dir = tempfile("tda"),
  ...
)

Arguments

data

a data frame or matrix, one row per case.

measure

the distance measure: 1 Euclidean (default), 2 city-block, 3 number of variables on which the two cases differ, 4 a dissimilarity index.

variables

optional subset of data's column names to use, instead of all of them.

weights

optional positive per-variable weights, one per (selected) variable, all distinct – see Details for TDA's parser constraint and the exact weighting convention.

options

a named list of further TDA options, passed through.

dir

working directory.

...

passed to tda_run. opt= (the command's variant switch) passes through as well.

Value

A dist object.

Details

As with tda_dma, TDA's v= needs variable names that start uppercase to be referenced correctly; variables handles this automatically.

Per-variable weights work through weights, with one genuine TDA quirk absorbed for you: wt= is parsed by TDA's time-points parser, which accepts only a strictly increasing positive list – wt=2,1,1 is a syntax error while wt=1,2,4 is fine, which is why this option long looked dead. Since a distance does not care about variable order, the wrapper sends the variables sorted by ascending weight. The weighting convention, verified against R directly: euclidean uses \(d^2 = \sum_j w_j \Delta_j^2\) (so it matches dist(sweep(x, 2, sqrt(w), "*"))), city-block \(d = \sum_j w_j |\Delta_j|\). Tied weights cannot pass TDA's parser at all; scale those columns yourself (\(\sqrt{w} x\) for euclidean, \(w x\) for city-block) and call this unweighted – the result is identical.

See also

Other clustering: TDA_CLUSTER, TDA_MDS, tda_conjoint(), tda_cutree(), tda_dma()

Examples

set.seed(1)
d <- data.frame(x1 = rnorm(5), x2 = rnorm(5))
tda_pdatd(d)
#>          V1       V2       V3       V4
#> V2 1.538458                           
#> V3 1.572765 1.049697                  
#> V4 2.624046 1.414400 2.436338         
#> V5 1.085896 0.806124 1.564251 1.542284
tda_pdatd(d, measure = 2)   # city-block, not the euclidean default
#>           V1        V2        V3        V4
#> V2 2.1179946                              
#> V3 1.7679679 1.2701676                    
#> V4 3.6179843 1.4999898 2.5934528          
#> V5 1.4710416 0.9386819 2.2088495 2.1469428
tda_pdatd(d, variables = "x1")   # a subset of the variables
#>           V1        V2        V3        V4
#> V2 0.8100971                              
#> V3 0.2091748 1.0192719                    
#> V4 2.2217346 1.4116375 2.4309094          
#> V5 0.9559616 0.1458644 1.1651364 1.2657730