Mahalanobis distance with missing entries

[thevolatilebit]

The Mahalanobis distance between a state with missing readouts for a subset of features and a "historical" row is essentially calculated as if we were setting the missing entry equal to the respective record in the "historical" row. See the following link for reference:

CEEDesigns.jl/src/GenerativeDesigns/distancebased.jl

Lines 115 to 128 in d1d09c3

	compute_distances = function (evidence::Evidence)
	if isempty(evidence)
	return zeros(nrow(data))
	else
	vec_evidence = map(colname -> get(evidence, colname, 0), non_targets)
	distances = map(eachrow(data)) do row
	vec_row = map(
	colname -> haskey(evidence, colname) ? row[colname] : 0,
	non_targets,
	)
	z = vec_evidence - vec_row
	dot(z, Λ * z)
	end

However, this might not be the optimal approach. For more information, consider the following resources: https://search.r-project.org/CRAN/refmans/modi/html/MDmiss.html and also https://www.jstor.org/stable/3559861, particularly on page 285.