dml: Distance Metric Learning in R

title: 'dml: Distance Metric Learning in R' authors: - affiliation: 1 name: Yuan Tang orcid: 0000-0001-5243-233X - affiliation: 2 name: Tao Gao - affiliation: 3 name: Nan Xiao orcid: 0000-0002-0250-5673 date: "19 October 2018" output: pdf_document: default html_document: df_print: paged bibliography: paper.bib tags: - R - distance metric learning - statistics - clustering - machine learning - dimensionality reduction affiliations: - index: 1 name: Ant Financial Services Group, San Mateo, CA 94402, USA - index: 2 name: Minghong Investment Management, Shanghai 200122, China - index: 3 name: Seven Bridges Genomics, Inc., Cambridge, MA 02142, USA

Summary

Distance metric is widely used in the machine learning literature. We used to choose a distance metric according to a priori (e.g. Euclidean Distance , L1 Distance, etc.) or according to the result of cross validation within small class of functions (e.g. choosing order of polynomial for a kernel). Actually, with priori knowledge of the data, we could learn a more suitable distance metric with (semi-)supervised distance metric learning techniques. dml [@dmlpkg] is such an R package aims to implement a collection of algorithms for (semi-)supervised distance metric learning.

The dml package provides native R implementations for a collection of Distance Metric Learning algorithms, including both global and local methods such as Relevant Component Analysis [@rcapaper], Discriminative Component Analysis [@dcapaper], and Local Fisher Discriminant Analysis [@lfdaoriginalpaper]. A list of all the implemented algorithms can be found in the dml package reference manual. These methods are widely applied in feature extraction, dimensionality reduction, clustering, information retrieval, and computer vision problems.

Additionally, implementations for the variants of the methods are also available in dml package. For example, since it was built on top of the lfda [@lfdapaper; @lfdapkg] package, users also have access to the family of Local Fisher Discriminant Analysis methods, which includes Local Fisher Discriminant Analysis, Kernel Local Fisher Discriminant Analysis, and Semi-supervised Local Fisher Discriminant Analysis [@semilfdapaper]. To make the results of these methods easy for users to interprete and analyze, both static and interactive visualizations for the results are available through ggfortify [@rjggfortify; @ggfortify] and autoplotly [@autoplotlypkg; @autoplotly2018joss] packages respectively.