Returns information about the distance measures that can be provided by name
to the dist_fun argument. The table describes their input
requirements, what they compare, the effect of rescaling individual raster
layers, and practical guidance.
Value
A data frame with one row per supported measure and columns:
nameValue accepted by
dist_fun.labelMeasure name.
inputType of the input vector.
comparesWhat aspect of the vectors the measure compares.
effect_of_layer_rescalingEffect of multiplying individual raster layers by positive constants.
guidancePractical usage guidance.
Details
Other measures can be supplied explicitly as a function to dist_fun.
In that case, the function must take two numeric vectors and return one
finite, nonnegative value, with smaller values meaning greater similarity.
For "cosine_distance", two zero profiles have distance zero. A zero
profile and a nonzero profile have distance one.
Examples
sc_list_distances()
#> name label
#> 1 euclidean Euclidean
#> 2 squared_euclidean Squared Euclidean
#> 3 manhattan Manhattan
#> 4 chebyshev Chebyshev
#> 5 canberra Canberra
#> 6 lorentzian Lorentzian
#> 7 soergel Soergel
#> 8 wavehedges Wave Hedges
#> 9 czekanowski Czekanowski
#> 10 squared_chord Squared chord
#> 11 squared_chi Squared chi-square
#> 12 clark Clark
#> 13 jsd Jensen-Shannon divergence
#> 14 non-intersection Non-intersection
#> 15 wasserstein1 Wasserstein-1
#> 16 cosine_distance Cosine distance
#> 17 dtw Dynamic time warping
#> 18 dtw2d Two-dimensional dynamic time warping
#> input
#> 1 numeric values
#> 2 numeric values
#> 3 numeric values
#> 4 numeric values
#> 5 numeric values
#> 6 numeric values
#> 7 nonnegative values
#> 8 nonnegative values
#> 9 nonnegative values
#> 10 nonnegative values
#> 11 nonnegative values
#> 12 nonnegative values
#> 13 probabilities summing to 1
#> 14 probabilities summing to 1
#> 15 probabilities in equal-width bins summing to 1
#> 16 numeric values
#> 17 ordered numeric sequence
#> 18 two equal-length coordinate sequences
#> compares effect_of_layer_rescaling
#> 1 root-sum-square layer differences can change distance
#> 2 summed squared layer differences can change distance
#> 3 summed absolute layer differences can change distance
#> 4 largest absolute layer difference can change distance
#> 5 relative layer differences no effect
#> 6 summed log-scaled layer differences can change distance
#> 7 difference relative to larger values can change distance
#> 8 summed relative layer differences no effect
#> 9 difference relative to total magnitude can change distance
#> 10 differences after square-root transformation can change distance
#> 11 squared differences relative to combined values can change distance
#> 12 combined relative layer differences no effect
#> 13 difference between probability profiles requires renormalization
#> 14 probability assigned differently across layers requires renormalization
#> 15 probability mass shifted across ordered bins requires renormalization
#> 16 pattern across layers, ignoring overall scale can change distance
#> 17 sequence difference after alignment can change distance
#> 18 2D trajectory difference after alignment can change distance
#> guidance
#> 1 Scale layers to comparable units or ranges.
#> 2 Scale layers to comparable units or ranges.
#> 3 Scale layers to comparable units or ranges.
#> 4 Scale layers to comparable units or ranges.
#> 5 Layer scaling has no effect.
#> 6 Scale layers to comparable units or ranges.
#> 7 Scale layers to set their influence.
#> 8 Layer scaling has no effect.
#> 9 Scale layers to set their influence.
#> 10 Scale layers to set their influence.
#> 11 Scale layers to set their influence.
#> 12 Layer scaling has no effect.
#> 13 Normalize every cell to sum to 1 and use mean centres.
#> 14 Normalize every cell to sum to 1 and use mean centres.
#> 15 Normalize every cell to sum to 1, use mean centres, and arrange layers in bin order.
#> 16 Scale layers only to change their influence.
#> 17 Arrange layers in sequence order.
#> 18 Arrange layers by dimension, with sequence order within each dimension.