Merge pull request #9 from JuliaAI/dev

For a 0.1.1 release

Author: ablaom

Project.toml

@@ -1,7 +1,7 @@
 name = "CategoricalDistributions"
 uuid = "af321ab8-2d2e-40a6-b165-3d674595d28e"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "0.1.0"
+version = "0.1.1"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
@@ -10,13 +10,15 @@ Missings = "e1d29d7a-bbdc-5cf2-9ac0-f12de2c33e28"
 OrderedCollections = "bac558e1-5e72-5ebc-8fee-abe8a469f55d"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 ScientificTypesBase = "30f210dd-8aff-4c5f-94ba-8e64358c1161"
+UnicodePlots = "b8865327-cd53-5732-bb35-84acbb429228"
 
 [compat]
 CategoricalArrays = "0.9, 0.10"
 Distributions = "0.25"
 Missings = "0.4, 1"
 OrderedCollections = "1.1"
 ScientificTypesBase = "2"
+UnicodePlots = "2"
 julia = "1.0"
 
 [extras]

README.md

@@ -1,14 +1,15 @@
 # CategoricalDistributions.jl
 
 Probability distributions and measures for finite sample spaces whose
-elements are *labeled*.
+elements are *labeled* (consist of the class pool of a
+`CategoricalArray`).
 
 Designed for performance in machine learning applications. For
 example, probabilistic classifiers in
 [MLJ](https://alan-turing-institute.github.io/MLJ.jl/dev/) typically
 predict the `UnivariateFiniteVector` objects defined in this package.
 
-For probability distributions over integers (unlabeled data) see the
+For probability distributions over integers see the
 [Distributions.jl](https://juliastats.org/Distributions.jl/stable/univariate/#Discrete-Distributions)
 package, whose methods the current package extends.
 
@@ -31,22 +32,16 @@ this package is the class pool of a `CategoricalArray`:
 ```julia
 using CategoricalDistributions
 using CategoricalArrays
-julia> data = rand(["yes", "no", "maybe"], 10) |> categorical
-10-element CategoricalArray{String,1,UInt32}:
- "maybe"
- "maybe"
- "no"
- "yes"
- "maybe"
- "no"
- "no"
- "no"
- "no"
- "yes"
-
-julia> d = fit(UnivariateFinite, data)
-UnivariateFinite{Multiclass{3}}(maybe=>0.3, no=>0.5, yes=>0.2)
-
+import Distributions
+data = ["no", "yes", "no", "maybe", "maybe", "no",
+       "maybe", "no", "maybe"] |> categorical
+julia> d = Distributions.fit(UnivariateFinite, data)
+               UnivariateFinite{Multiclass{3}}
+         ┌                                        ┐
+   maybe ┤■■■■■■■■■■■■■■■■■■■■■■■■■■■■ 0.4
+      no ┤■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■ 0.5
+     yes ┤■■■■■■■ 0.1
+         └                                        ┘
 julia> pdf(d, "no")
 0.5
 
@@ -59,7 +54,11 @@ from a probability vector:
 
 ```julia
 julia> d2 = UnivariateFinite(["no", "yes"], [0.15, 0.85], pool=data)
-UnivariateFinite{Multiclass{3}}(no=>0.15, yes=>0.85)
+             UnivariateFinite{Multiclass{3}}
+       ┌                                        ┐
+    no ┤■■■■■■ 0.15
+   yes ┤■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■ 0.85
+       └                                        ┘
 ```
 
 A `UnivariateFinite` distribution tracks all classes in the pool:
@@ -74,12 +73,12 @@ levels(d2)
 julia> pdf(d2, "maybe")
 0.0
 
-julia> pdf(d2, "okay")
+julia> pdf(d2, "okay")https://github.com/JuliaAI/CategoricalDistributions.jl#measures-over-finite-labeled-sets
 ERROR: DomainError with Value okay not in pool. :
 ```
 
 Arrays of `UnivariateFinite` distributions are defined using the same
-constructor. Broadcasting methods, such as `pdf`, is optimized for
+constructor. Broadcasting methods, such as `pdf`, are optimized for
 such arrays:
 
 ```
@@ -148,8 +147,8 @@ over finite labeled sets.
   Distributions.jl, with efficient broadcasting over the new array
   type.
 
-- Implementation of `fit` from Distributions.jl for `UnivariateFinite`
-  distributions.
+- Implementation of `Distributions.fit` from Distributions.jl for
+  `UnivariateFinite` distributions.
 
 - A single constructor for constructing `UnivariateFinite`
     distributions and arrays thereof, from arrays of probabilities.

src/CategoricalDistributions.jl

@@ -1,19 +1,14 @@
 module CategoricalDistributions
 
-export UnivariateFinite, UnivariateFiniteArray
-
-# re-eported from Distributions:
-export pdf, logpdf, support, mode
-
 import Distributions
-import ScientificTypesBase
+import ScientificTypesBase: Finite, Multiclass, OrderedFactor
 using OrderedCollections
 using CategoricalArrays
 import Missings
 using Random
+using UnicodePlots
 
 const Dist = Distributions
-const STB = ScientificTypesBase
 
 import Distributions: pdf, logpdf, support, mode
 
@@ -22,4 +17,12 @@ include("types.jl")
 include("methods.jl")
 include("arrays.jl")
 
+export UnivariateFinite, UnivariateFiniteArray
+
+# re-eport from Distributions:
+export pdf, logpdf, support, mode
+
+# re-export from ScientificTypesBase:
+export Multiclass, OrderedFactor
+
 end

src/methods.jl

@@ -79,6 +79,15 @@ function Base.show(stream::IO, d::UnivariateFinite)
     print(stream, "UnivariateFinite{$(d.scitype)}($arg_str)")
 end
 
+function Base.show(io::IO, mime::MIME"text/plain",
+                   d::UnivariateFinite{S}) where S
+    s = support(d)
+    x = string.(CategoricalArrays.DataAPI.unwrap.(s))
+    y = pdf.(d, s)
+    plt = barplot(x, y, title="UnivariateFinite{$S}")
+    show(io, mime, plt)
+end
+
 show_prefix(u::UnivariateFiniteArray{S,V,R,P,1}) where {S,V,R,P} =
     "$(length(u))-element"
 show_prefix(u::UnivariateFiniteArray) = join(size(u),'x')

src/types.jl

@@ -131,7 +131,7 @@ same size as the array.
 
 # extend Ditributions type hiearchy to account for non-euclidean
 # supports:
-abstract type Categorical{S<:STB.Finite} <: Dist.ValueSupport end
+abstract type Categorical{S<:Finite} <: Dist.ValueSupport end
 
 # not exported:
 const _UnivariateFinite_{S} =

src/utilities.jl

@@ -4,7 +4,7 @@
 
 function scitype(c::CategoricalValue)
     nc = length(levels(c.pool))
-    return ifelse(c.pool.ordered, STB.OrderedFactor{nc}, STB.Multiclass{nc})
+    return ifelse(c.pool.ordered, OrderedFactor{nc}, Multiclass{nc})
 end