MDSReferenceMap: Construction of a MDS Reference Map
In tchitchek-lab/MDSRefMaps: What the Package Does in One 'Title Case' Line

Description Usage Arguments Details Value

This function computes a MDS Reference Map based on a distance matrix.

A MDS Reference Map corresponds to a regular MDS representation on which additional objects can be projected. MDS Reference Maps can be computed based on the Euclidean or Manhattan metrics using the 'metric' parameter. The initialization space of object positions can be specified using the 'init' parameter.

1
2
3

MDSReferenceMap(dist, k = 2, init = "svd", metric = "euclidean",
  max_it = 6 * 10^6, stress_sd_th = 10^-4, stack_length = 500,
  verbose = TRUE)

`dist`	a numeric matrix with all pairwise distances between objects of the representation
`k`	a numeric value specifying the desired number of dimensions in the resulting Reference Map representation
`init`	a character value or a numeric matrix specifying how the objects are positioned in the initial configuration. Possible character values are 'rand', 'center', 'svd' (please refer to the details section for more details). Object positions in the initial configuration can be explicitly specified using a numeric matrix where the rows correspond to the objects and where the columns correspond to the MDS dimensions (in k dimensions).
`metric`	a character indicating the distance metric to use ("euclidean" or "manhattan")
`max_it`	a numeric defining the maximal number of steps the algorithm can perform
`stress_sd_th`	a numeric defining the threshold for the standard deviation of Kruskal Stress
`stack_length`	a numeric defining the length of the Kruskal Stress stack (used to compute the standard deviation of the Kruskal Stress)
`verbose`	a boolean enabling the display of debug information at each step of the algorithm

The RefMaps algorithm implements SVD-MDS algorithm which is based on a molecular dynamic approach (Becavin et al.). This metric performs a dimensionality reduction of the original space by modeling objects by particles and pairwise distances between them by repulsion and attraction forces. SVD-MDS metric use Verlet algorithm (Loup Verlet in 1967) to compute the MDS representation. Algorithm constants can be specified via the 'setConts()' function.

This implementation allows to used incomplete distance matrices (distance matrices with missing values modeled by NA). Furthermore, distance matrices can be computed based on the Euclidean or Manhattan metrics.

This implementation has been implemented in C++ to handle large sets of high-dimensional objects. init. KS+entourage.

a list of 3 elements containing the position of the objects ('points' element), the Kruskal Stress ('stress' element), and the Entourage Score ('entourage' element)

tchitchek-lab/MDSRefMaps documentation built on May 31, 2019, 7:28 a.m.