Merge branch 'master' into jpata/abstract-hist

REQUIRE

@@ -1,2 +1,2 @@
 julia 0.3-
-ArrayViews 0.4.6-
+ArrayViews 0.4.8-

docs/source/scalarstats.rst

@@ -67,10 +67,12 @@ Measurements of Variation
 
   Compute the standard error of the mean for ``x``, *i.e.* ``sqrt(var(x) / length(x))``.
 
-.. function:: mad(x)
+.. function:: mad(x[, center][; constant=1.4826])
 
   Compute the `median absolute deviation <http://en.wikipedia.org/wiki/Median_absolute_deviation>`_ of ``x``.
 
+  One can optionally supply the ``center``. By default, ``constant=1.4826`` for consistent estimation of the standard deviation of a normal distribution.
+
 
 Z-scores
 ----------
@@ -139,7 +141,7 @@ Quantile and Friends
 
   Extended method of *quantile*. Equivalent to ``nquantile(x, 4)``, which returns a vector of quantiles at ``0.0, 0.25, 0.50, 0.75, 1.0``. 
 
-.. function:: median(x, w[; checknan=true])
+.. function:: median(x, w)
 
   Compute the weighted median of ``x``, using weights given by a weight vector ``w`` (of type ``WeightVec``).  The weight and data vectors must have the same length.  The weighted median :math:`x_k` is the element of ``x`` that satisfies :math:`\sum_{x_i < x_k} w_i \le \frac{1}{2} \sum_{j} w_j` and :math:`\sum_{x_i > x_k} w_i \le \frac{1}{2} \sum_{j} w_j`.  If a weight has value zero, then its associated data point is ignored.  If none of the weights are positive, an error is thrown.  ``NaN`` is returned if ``x`` contains any ``NaN`` values.  An error is raised if ``w`` contains any ``NaN`` values.
 

src/StatsBase.jl

@@ -3,10 +3,20 @@ module StatsBase
 
     import Base: length, isempty, eltype, values, sum, mean, mean!, show, quantile
     import Base: rand, rand!
-    import Base: Func, evaluate, IdFun, Abs2Fun
+    import Base: Func, IdFun, Abs2Fun
     import Base.LinAlg: BlasReal, BlasFloat
     import Base.Cartesian: @ngenerate, @nloops, @nref, @nextract
 
+    ## tackle compatibility issues
+
+    if VERSION < v"0.4.0-dev+1258"
+        import Base: evaluate
+    else
+        evaluate(f::Func{1}, x) = call(f, x)
+        evaluate(f::Func{2}, x, y) = call(f, x, y)
+    end
+
+
     export
 
     # reexport from ArrayViews

src/misc.jl

@@ -112,8 +112,7 @@ function indicatormat(x::AbstractArray, c::AbstractArray; sparse::Bool=false)
     sparse ? _indicatormat_sparse(x, c) : _indicatormat_dense(x, c)
 end
 
-indicatormat{T<:Union(Real,String)}(x::AbstractArray{T}; sparse::Bool=false) = indicatormat(x, sort!(unique(x)); sparse=sparse)
-indicatormat(x::AbstractArray; sparse::Bool=false) = indicatormat(x, unique(x); sparse=sparse)
+indicatormat(x::AbstractArray; sparse::Bool=false) = indicatormat(x, sort!(unique(x)); sparse=sparse)
 
 
 function _indicatormat_dense(x::IntegerArray, k::Integer)

src/rankcorr.jl

@@ -88,11 +88,11 @@ end
 
 corkendall(x::RealVector, y::RealVector) = corkendall!(float(copy(x)), float(copy(y)))
 
-corkendall(X::RealMatrix, y::RealVector) = [corkendall!(float(X[:,i]), float(copy(y))) for i in 1:size(X, 2)]
+corkendall(X::RealMatrix, y::RealVector) = Float64[corkendall!(float(X[:,i]), float(copy(y))) for i in 1:size(X, 2)]
 
-corkendall(x::RealVector, Y::RealMatrix) = (n = size(Y,2); reshape([corkendall!(float(copy(x)), float(Y[:,i])) for i in 1:n], 1, n))
+corkendall(x::RealVector, Y::RealMatrix) = (n = size(Y,2); reshape(Float64[corkendall!(float(copy(x)), float(Y[:,i])) for i in 1:n], 1, n))
 
-corkendall(X::RealMatrix, Y::RealMatrix) = [corkendall!(float(X[:,i]), float(Y[:,j])) for i in 1:size(X, 2), j in 1:size(Y, 2)]
+corkendall(X::RealMatrix, Y::RealMatrix) = Float64[corkendall!(float(X[:,i]), float(Y[:,j])) for i in 1:size(X, 2), j in 1:size(Y, 2)]
 
 function corkendall(X::RealMatrix)
     n = size(X, 2)

src/scalarstats.jl

@@ -173,20 +173,16 @@ variation{T<:Real}(x::AbstractArray{T}) = variation(x, mean(x))
 sem{T<:Real}(a::AbstractArray{T}) = sqrt(var(a) / length(a))
 
 # Median absolute deviation
-mad{T<:Real}(v::AbstractArray{T}, center::Real) = 1.4826 * median!(abs(v .- center))
+mad{T<:Real}(v::AbstractArray{T}, args...;arg...) = mad!(copy(v), args...;arg...)
+mad{T<:Real}(v::Range{T}, args...;arg...) = mad!([v], args...;arg...)
 
-function mad!{T<:Real}(v::AbstractArray{T}, center::Real)
+function mad!{T<:Real}(v::AbstractArray{T}, center::Real=median!(v); constant::Real=1.4826)
     for i in 1:length(v)
-        v[i] = abs(v[i]-center)
+        @inbounds v[i] = abs(v[i]-center)
     end
-    1.4826 * median!(v)
+    constant * median!(v)
 end
 
-mad!{T<:Real}(v::AbstractArray{T}) = mad!(v, median!(v))
-
-mad{T<:Real}(v::AbstractArray{T}) = mad!(copy(v))
-mad{T<:Real}(v::Range{T}) = mad!([v])
-
 # Interquartile range
 iqr{T<:Real}(v::AbstractArray{T}) = (q = quantile(v, [.25, .75]); q[2] - q[1])
 

test/rankcorr.jl

@@ -3,7 +3,7 @@
 using StatsBase
 using Base.Test
 
-X = [1 0; 2 1; 3 0; 4 1; 5 10]
+X = Float64[1 0; 2 1; 3 0; 4 1; 5 10]
 
 x1 = X[:,1]
 x2 = X[:,2]

test/scalarstats.jl

@@ -76,6 +76,8 @@ z2 = [8. 2. 3. 1.; 24. 10. -1. -1.; 20. 12. 1. -2.]
 @test_approx_eq mad([1:5], 3) 1.4826
 @test_approx_eq mad([1:5]) 1.4826
 @test_approx_eq mad(1:5) 1.4826
+@test_approx_eq mad([1:5], constant=1.0) 1.0
+@test_approx_eq mad([1:5], 3, constant=1.0) 1.0
 
 @test_approx_eq iqr(1:5) 2.0