Merge pull request #23 from JuliaAI/dev

For a 0.1.5 release

Author: ablaom

Project.toml

@@ -1,7 +1,7 @@
 name = "StatisticalMeasures"
 uuid = "a19d573c-0a75-4610-95b3-7071388c7541"
 authors = ["Anthony D. Blaom <anthony.blaom@gmail.com>"]
-version = "0.1.4"
+version = "0.1.5"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"

src/confusion_matrices.jl

@@ -185,8 +185,8 @@ ordered confusion matrices, see [`$CM.confmat`](@ref).
 
 """
 struct ConfusionMatrix{N,O,L}
-    mat::Matrix{Int}
-    index_given_level::LittleDict{L, Int, NTuple{N,L}, NTuple{N,Int}}
+    mat::Matrix{<:Integer}
+    index_given_level::LittleDict{L, I, NTuple{N,L}, NTuple{N,I}} where I<:Integer
 end
 
 """
@@ -204,14 +204,13 @@ See also [`$CM.confmat`](@ref).
 """
 function ConfusionMatrix(
     m,
-    dic::AbstractDict{L,I};
+    dic::LittleDict{L, I, NTuple{N,L}, NTuple{N,I}};
     checks=true,
     ordered=false,
-    ) where {L,I<:Integer}
-    s = size(m)
-    N = s[1]
+    ) where {L,I<:Integer,N}
     if checks
-        N == s[2] || throw(ArgumentError("Expected a square matrix."))
+        s = size(m)
+        N == s[1] == s[2] || throw(ArgumentError("Expected a square matrix."))
         N > 1 || throw(ArgumentError("Expected a matrix of size ≥ 2x2."))
         length(unique(keys(dic))) == N || throw(ArgumentError(
         "Expected dictionary with $N unique keys (levels) as "*
@@ -222,8 +221,15 @@ function ConfusionMatrix(
             "to be integers from 1 to $N. "
         ))
     end
-    index_given_level = freeze(dic)
-    ConfusionMatrix{N,ordered,L}(m, index_given_level)
+    ConfusionMatrix{N,ordered,L}(m, dic)
+end
+function ConfusionMatrix(
+    m,
+    dic::AbstractDict;
+    checks=true,
+    ordered=false,
+    ) 
+    ConfusionMatrix(m, freeze(dic); checks, ordered)
 end
 
 """
@@ -251,7 +257,7 @@ function ConfusionMatrix(m, levels::AbstractVector{L}; ordered=false, checks=tru
         ))
     end
     index_given_level =
-        LittleDict{L, Int, Vector{L}, Vector{Int}}(levels, eachindex(levels)) |> freeze
+        LittleDict{L, Int, Vector{L}, Vector{Int}}(levels, eachindex(levels))
     ConfusionMatrix(m, index_given_level; ordered, checks=false)
 end
 
@@ -489,8 +495,8 @@ function confmat(ŷ, y, _levels, _perm, rev)
     perm = permutation(_perm,  rev, levels)
 
     levels = apply(perm, levels)
-    indexer = LittleDict(levels[i] => i for i in eachindex(levels)) |> freeze
-
+    L = eltype(levels)
+    indexer = LittleDict{L, Int, Vector{L}, Vector{Int}}(levels, eachindex(levels)) |> freeze
     _confmat(ŷ, y, indexer, levels, ordered)
 end
 
@@ -533,18 +539,26 @@ end
 
 
 # ## Final method to do the computation
-
-function _confmat(ŷ, y, indexer, levels, ordered)
+function _confmat(ŷ, y, indexer::F, levels, ordered) where F
     nc = length(levels)
     cmat = zeros(Int, nc, nc)
     @inbounds for i in eachindex(y)
         (ismissing(y[i]) || ismissing(ŷ[i])) && continue
         cmat[get(indexer, ŷ[i]), get(indexer, y[i])] += 1
     end
-    index_given_level = LittleDict(c => get(indexer, c) for c in levels) |> freeze
     return ConfusionMatrix(cmat, levels; ordered, checks=false)
 end
 
+function _confmat(ŷ, y, indexer::AbstractDict{L,I}, levels, ordered) where {L,I<:Integer}
+    nc = length(levels)
+    cmat = zeros(Int, nc, nc)
+    @inbounds for i in eachindex(y)
+        (ismissing(y[i]) || ismissing(ŷ[i])) && continue
+        cmat[get(indexer, ŷ[i]), get(indexer, y[i])] += 1
+    end
+    return ConfusionMatrix(cmat, indexer; ordered, checks=false)
+end
+
 
 # DISPLAY
 

src/continuous.jl

@@ -351,10 +351,10 @@ const RootMeanSquaredProportionalErrorDoc = docstring(
     scitype=DOC_INFINITE,
     body=
 """
-Specifically, compute the mean of ``(ŷ_i-y_i \\over y_i)^2}`` over all pairs of
-observations ``(ŷ_i, y_i)`` in `(ŷ, y)`, and return the square root of the result. More
+Specifically, compute the mean of `((ŷᵢ-yᵢ)/yᵢ)^2}` over all pairs of
+observations `(ŷᵢ, yᵢ)` in `(ŷ, y)`, and return the square root of the result. More
 generally, pre-multiply the values averaged by the specified weights. Terms for which
-``|y_i|``<`tol` are dropped in the summation, but counts still
+`abs(yᵢ) < tol` are dropped in the summation, but counts still
 contribute to the mean normalization factor.
 """,
 )

src/finite.jl

@@ -293,7 +293,13 @@ const BalancedAccuracyType = API.FussyMeasure{
     human_name="balanced accuracy",
 )
 
-register(BalancedAccuracy, "balanced_accuracy", "bacc", "bac")
+register(
+    BalancedAccuracy,
+    "balanced_accuracy",
+    "bacc",
+    "bac",
+    "Ppprobability_of_correct_classification",
+)
 
 const BalancedAccuracyDoc = docstring(
     "BalancedAccuracy(; adjusted=false)",

src/probabilistic.jl

@@ -178,7 +178,7 @@ const LogScoreDoc = docstring(
     body=
 """
 The score is a mean of observational scores. More generally, observational scores are
-pre-multiplied by the specified weights before avaraging. See below for the form that
+pre-multiplied by the specified weights before averaging. See below for the form that
 probabilistic predictions `ŷ` should take. Raw probabilities are clamped away from `0` and
 `1`. Specifically, if `p` is the probability mass/density function evaluated at given
 observed ground truth observation `η`, then the score for that example is defined as
@@ -228,6 +228,7 @@ const LogLossType = API.FussyMeasure{<:API.RobustMeasure{<:_LogLossType}}
 @trait(
     _LogLossType,
     consumes_multiple_observations=true,
+    can_report_unaggregated=true,
     kind_of_proxy=LearnAPI.Distribution(),
     # observation_scitype depends on distribution type
     observation_scitype = Union{Missing,Finite,Infinite},
@@ -328,7 +329,7 @@ const BrierScoreDoc = docstring(
     body=
 """
 The score is a mean of observational scores. More generally, observational scores are
-pre-multiplied by the specified weights before avaraging. See below for the form that
+pre-multiplied by the specified weights before averaging. See below for the form that
 probabilistic predictions `ŷ` should take.
 
 Convention as in $PROPER_SCORING_RULES
@@ -398,6 +399,7 @@ const BrierLossType = API.FussyMeasure{<:API.RobustMeasure{<:_BrierLossType}}
 @trait(
     _BrierLossType,
     consumes_multiple_observations=true,
+    can_report_unaggregated=true,
     kind_of_proxy=LearnAPI.Distribution(),
     # observation_scitype depends on distribution type
     observation_scitype = Union{Missing,Finite,Infinite},
@@ -513,7 +515,7 @@ const SphericalScoreDoc = docstring(
     body=
 """
 The score is a mean of observational scores. More generally, observational scores are
-pre-multiplied by the specified weights before avaraging. See below for the form that
+pre-multiplied by the specified weights before averaging. See below for the form that
 probabilistic predictions `ŷ` should take.
 
 Convention as in $PROPER_SCORING_RULES: If `y` takes on a finite