Edit on GitHub

VisualStringDistances

VisualStringDistances.Glyph — Method
Glyph(s::AbstractString) --> Glyph

Construct a Glyph from a string.

Examples

julia> Glyph("abc")
------------------------
------------------------
------------------------
---------#--------------
---------#--------------
---------#--------------
--####---#-###----####--
-#----#--##---#--#----#-
------#--#----#--#------
--#####--#----#--#------
-#----#--#----#--#------
-#----#--#----#--#------
-#---##--##---#--#----#-
--###-#--#-###----####--
------------------------
------------------------
source
VisualStringDistances.glyph! — Method
glyph!(v::Vector{UInt8}) -> Glyph

Creates a Glyph for a vector of bytes, assuming the vector represents a single Unifont character. Modifies v and may share its memory.

source
VisualStringDistances.visual_distance — Method
visual_distance(::Type{T}, s::Union{Char,AbstractString},
                     t::Union{Char,AbstractString}; D=KL(one(T)), ϵ=T(0.1),
                     normalize=nothing) where {T}

Computes a measure of distance between the strings s and t in terms of their visual representation as rendered by GNU Unifont and quantified by an unbalanced Sinkhorn divergence from UnbalancedOptimalTransport.jl.

  • The keyword argument D chooses the UnbalancedOptimalTransport.AbstractDivergence used to penalize the creation or destruction of "mass" (black pixels). For D = VisualStringDistances.KL(ρ) for some number ρ ≥ 0, the distance is non-negative and zero if and only if the two visual representations of the strings are the same, as is generally desired.
  • The keyword argument ϵ sets the "entropic regularization" in the Sinkhorn divergence; see the documentation there for more information. In short, smaller ϵ computes a quantity more directly related to the cost of moving mass, but takes longer to compute.
  • The keyword argument normalize can be chosen to be a function which returns a normalizing constant given the maximum length of the two strings. The choice normalize=identity thus divides the result by the maximum length of the two strings. The choice normalize=sqrt has been found to give a good balance in some settings.

One may use printglyph to see the visual representation of the strings as rendered by GNU Unifont.

Note

At the time of this writing, GNU Unifont is capable of rendering 57086 different unicode characters. However, it renders some unicode characters with the same graphical representation; specifically, 689 distinct unicode characters have duplicate representations. Here's a set of six duplicates, for example:

  • 'Ꮋ': Unicode U+13BB (category Lu: Letter, uppercase)
  • 'Н': Unicode U+041D (category Lu: Letter, uppercase)
  • 'ꓧ': Unicode U+A4E7 (category Lo: Letter, other)
  • 'Ⲏ': Unicode U+2C8E (category Lu: Letter, uppercase)
  • 'Η': Unicode U+0397 (category Lu: Letter, uppercase)
  • 'H': ASCII/Unicode U+0048 (category Lu: Letter, uppercase)

The visual distance between these, therefore, is returned as zero (up to numerical error).

Example

julia> using VisualStringDistances

julia> printglyph("abc")
------------------------
------------------------
------------------------
---------#--------------
---------#--------------
---------#--------------
--####---#-###----####--
-#----#--##---#--#----#-
------#--#----#--#------
--#####--#----#--#------
-#----#--#----#--#------
-#----#--#----#--#------
-#---##--##---#--#----#-
--###-#--#-###----####--
------------------------
------------------------

julia> printglyph("def")
------------------------
------------------------
------------------------
------#-------------##--
------#------------#----
------#------------#----
--###-#---####-----#----
-#---##--#----#--#####--
-#----#--#----#----#----
-#----#--######----#----
-#----#--#---------#----
-#----#--#---------#----
-#---##--#----#----#----
--###-#---####-----#----
------------------------
------------------------

julia> visual_distance("abc", "def")
31.57060117541754

julia> visual_distance("abc", "abe")
4.979840716647487
source