Changes for Julia 0.7/1.0 (JuliaStats#45)

* Modified code to support Julia 0.7 - 1.0

* added dependencies

* Removed Iterators

* Added version check for initial imports

* Fixed deprecation warning (Iterators)

* Addressed commets raised by ararslan

* Removed VERSION checks and code for < v"0.7.0"

* Update .travis.yml

Co-authored-by: Anand Bisen <[email protected]>

.travis.yml

@@ -3,7 +3,8 @@ os:
   - linux
   - osx
 julia:
-  - 0.6
+  - 0.7
+  - 1.0
   - nightly
 notifications:
   email: false
@@ -12,4 +13,4 @@ notifications:
 #   - if [[ -a .git/shallow ]]; then git fetch --unshallow; fi
 #   - julia -e 'Pkg.clone(pwd()); Pkg.build("MLBase");  Pkg.test("MLBase")'
 after_success:
-  - julia -e 'cd(Pkg.dir("MLBase")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())'
+  - julia -e 'using Pkg; Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())'

REQUIRE

@@ -1,4 +1,4 @@
-julia 0.6
+julia 0.7
 Reexport
 StatsBase 0.6.9-
 IterTools

src/MLBase.jl

@@ -2,12 +2,14 @@ module MLBase
 
     using Reexport
     using IterTools
+    using Random
     @reexport using StatsBase
 
     import Base: length, show, keys, precision, length, getindex
-    import Base: start, next, done
+    import Base: iterate
     import Base.Order: lt, Ordering, ForwardOrdering, ReverseOrdering, Forward, Reverse
     import StatsBase: RealVector, IntegerVector, RealMatrix, IntegerMatrix, RealArray
+    import IterTools: product
 
     export
 
@@ -77,4 +79,3 @@ module MLBase
 
     include("deprecates.jl")
 end
-

src/classification.jl

@@ -32,7 +32,8 @@ end
 
 classify!(r::IntegerVector, x::RealMatrix) = classify!(r, x, Forward)
 
-classify(x::RealMatrix, ord::Ordering) = classify!(Array{Int}(size(x,2)), x, ord)
+# - this one throws a deprecation
+classify(x::RealMatrix, ord::Ordering) = classify!(Array{Int}(undef, size(x,2)), x, ord)
 classify(x::RealMatrix) = classify(x, Forward)
 
 # classify with score(s)
@@ -65,13 +66,13 @@ function classify_withscores!(r::IntegerVector, s::RealVector, x::RealMatrix, or
     return (r, s)
 end
 
-classify_withscores!(r::IntegerVector, s::RealVector, x::RealMatrix) = 
+classify_withscores!(r::IntegerVector, s::RealVector, x::RealMatrix) =
     classify_withscores!(r, s, x, Forward)
 
 function classify_withscores(x::RealMatrix{T}, ord::Ordering) where T<:Real
     n = size(x, 2)
-    r = Array{Int}(n)
-    s = Array{T}(n)
+    r = Array{Int}(undef, n)
+    s = Array{T}(undef, n)
     return classify_withscores!(r, s, x, ord)
 end
 
@@ -80,7 +81,7 @@ classify_withscores(x::RealMatrix{T}) where {T<:Real} = classify_withscores(x, F
 
 # classify with threshold
 
-classify(x::RealVector, t::Real, ord::Ordering) = 
+classify(x::RealVector, t::Real, ord::Ordering) =
     ((k, v) = classify_withscore(x, ord); ifelse(lt(ord, v, t), 0, k))
 
 classify(x::RealVector, t::Real) = classify(x, t, Forward)
@@ -97,8 +98,8 @@ end
 
 classify!(r::IntegerVector, x::RealMatrix, t::Real) = classify!(r, x, t, Forward)
 
-classify(x::RealMatrix, t::Real, ord::Ordering) = classify!(Array{Int}(size(x,2)), x, t, ord)
-classify(x::RealMatrix, t::Real) = classify(x, t, Forward)  
+classify(x::RealMatrix, t::Real, ord::Ordering) = classify!(Array{Int}(undef, size(x,2)), x, t, ord)
+classify(x::RealMatrix, t::Real) = classify(x, t, Forward)
 
 
 ## label map
@@ -109,7 +110,7 @@ struct LabelMap{K}
 
     function LabelMap{K}(vs, v2i) where K
         length(vs) == length(v2i) || throw(DimensionMismatch("lengths of vs and v2i mismatch"))
-        new(vs,v2i) 
+        new(vs,v2i)
     end
 end
 
@@ -143,18 +144,18 @@ end
 
 # use a map to encode discrete values into labels
 labelencode(lmap::LabelMap{T}, x) where {T} = lmap.v2i[convert(T, x)]
-labelencode(lmap::LabelMap{T}, xs::AbstractArray{T}) where {T} = 
+labelencode(lmap::LabelMap{T}, xs::AbstractArray{T}) where {T} =
     reshape(Int[labelencode(lmap, x) for x in xs], size(xs))
 
 # decode the label to the associated discrete value
 labeldecode(lmap::LabelMap{T}, y::Int) where {T} = lmap.vs[y]
-labeldecode(lmap::LabelMap{T}, ys::AbstractArray{Int}) where {T} = 
+labeldecode(lmap::LabelMap{T}, ys::AbstractArray{Int}) where {T} =
     reshape(T[labeldecode(lmap, y) for y in ys], size(ys))
 
 ## group labels
 
 function groupindices(k::Int, xs::IntegerVector; warning::Bool=true)
-    gs = Array{Vector{Int}}(k)
+    gs = Array{Vector{Int}}(undef, k)
     for i = 1:k
         gs[i] = Int[]
     end
@@ -176,7 +177,7 @@ end
 
 function groupindices(lmap::LabelMap{T}, xs::AbstractArray{T}) where T
     k = length(lmap)
-    gs = Array{Vector{Int}}(k)
+    gs = Array{Vector{Int}}(undef, k)
     for i = 1:k
         gs[i] = Int[]
     end
@@ -187,4 +188,3 @@ function groupindices(lmap::LabelMap{T}, xs::AbstractArray{T}) where T
     end
     return gs
 end
-

src/crossval.jl

@@ -4,7 +4,7 @@
 
 abstract type CrossValGenerator end
 
-# K-fold 
+# K-fold
 
 struct Kfold <: CrossValGenerator
     permseq::Vector{Int}
@@ -20,15 +20,20 @@ end
 length(c::Kfold) = c.k
 
 struct KfoldState
-    i::Int      # the i-th of the subset 
+    i::Int      # the i-th of the subset
     s::Int      # starting index
     e::Int      # ending index
 end
 
-start(c::Kfold) = KfoldState(1, 1, round.(Integer,c.coeff))
-next(c::Kfold, s::KfoldState) =
-    (i = s.i+1; (setdiff(1:length(c.permseq), c.permseq[s.s:s.e]), KfoldState(i, s.e+1, round.(Integer,c.coeff * i))))
-done(c::Kfold, s::KfoldState) = (s.i > c.k)
+# A version check allows to maintain compatibility with earlier versions
+function Base.iterate(c::Kfold, state::KfoldState=KfoldState(1, 1, round.(Integer, c.coeff)))
+    i, s, e = state.i, state.s, state.e
+    (i > c.k) && return nothing
+    i += 1
+    sd = setdiff(1:length(c.permseq), c.permseq[s:e])
+    kst = KfoldState(i, e + 1, round.(Integer, c.coeff * i))
+    return (i; (sd, kst))
+end
 
 # Stratified K-fold
 
@@ -52,22 +57,21 @@ end
 
 length(c::StratifiedKfold) = c.k
 
-start(c::StratifiedKfold) = 1
-function next(c::StratifiedKfold, s::Int)
+function Base.iterate(c::StratifiedKfold, s::Int=1)
+    (s > c.k) && return nothing
     r = Int[]
     for (permseq, coeff) in zip(c.permseqs, c.coeffs)
         a, b = round.(Integer, [s-1, s] .* coeff)
         append!(r, view(permseq, a+1:b))
     end
-    setdiff(1:c.n, r), s+1
+    return setdiff(1:c.n, r), s+1
 end
-done(c::StratifiedKfold, s::Int) = (s > c.k)
 
 # LOOCV (Leave-one-out cross-validation)
 
 function leave_one_out(n::Int, i::Int)
     @assert 1 <= i <= n
-    x = Array{Int}(n-1)
+    x = Array{Int}(undef, n - 1)
     for j = 1:i-1
         x[j] = j
     end
@@ -83,10 +87,10 @@ end
 
 length(c::LOOCV) = c.n
 
-start(c::LOOCV) = 1
-next(c::LOOCV, s::Int) = (leave_one_out(c.n, s), s+1)
-done(c::LOOCV, s::Int) = (s > c.n)
-
+function iterate(c::LOOCV, s::Int=1)
+    (s > c.n) && return nothing
+    return (leave_one_out(c.n, s), s + 1)
+end
 
 # Repeated random sub-sampling
 
@@ -98,9 +102,10 @@ end
 
 length(c::RandomSub) = c.k
 
-start(c::RandomSub) = 1
-next(c::RandomSub, s::Int) = (sort!(sample(1:c.n, c.sn; replace=false)), s+1)
-done(c::RandomSub, s::Int) = (s > c.k)
+function iterate(c::RandomSub, s::Int=1)
+    (s > c.k) && return nothing
+    return (sort!(sample(1:c.n, c.sn; replace=false)), s+1)
+end
 
 # Stratified repeated random sub-sampling
 
@@ -134,24 +139,23 @@ end
 
 length(c::StratifiedRandomSub) = c.k
 
-start(c::StratifiedRandomSub) = 1
-function next(c::StratifiedRandomSub, s::Int)
-    idxs = Array{Int}(0)
+function iterate(c::StratifiedRandomSub, s::Int=1)
+    (s > c.k) && return nothing
+    idxs = Array{Int}(undef, 0)
     sizehint!(idxs, c.sn)
     for (stratum_sn, stratum_idxs) in zip(c.sns, c.idxs)
         append!(idxs, sample(stratum_idxs, stratum_sn, replace=false))
     end
-    (sort!(idxs), s+1)
+    return (sort!(idxs), s + 1)
 end
-done(c::StratifiedRandomSub, s::Int) = (s > c.k)
 
 ## Cross validation algorithm
 #
 #  estfun: model estimation function
 #
 #          model = estfun(train_inds)
 #
-#          it takes as input the indices of 
+#          it takes as input the indices of
 #          the samples for training, and returns
 #          a trained model.
 #
@@ -172,7 +176,7 @@ done(c::StratifiedRandomSub, s::Int) = (s > c.k)
 #
 function cross_validate(estfun::Function, evalfun::Function, n::Int, gen)
     best_model = nothing
-    best_score = NaN   
+    best_score = NaN
     best_inds = Int[]
     first = true
 
@@ -186,8 +190,5 @@ function cross_validate(estfun::Function, evalfun::Function, n::Int, gen)
     return scores
 end
 
-cross_validate(estfun::Function, evalfun::Function, n::Integer, gen) = 
+cross_validate(estfun::Function, evalfun::Function, n::Integer, gen) =
     cross_validate(estfun, evalfun, n, gen, Forward)
-
-
-

src/modeltune.jl

@@ -16,7 +16,7 @@ function gridtune(estfun::Function,                          # model estimation
     t = 0
     first = true
     local best_score, best_model, best_cfg
-    for cf in product(pvals...)
+    for cf in Iterators.product(pvals...)
         t += 1
         m = estfun(cf...)
         if m == nothing

src/perfeval.jl

@@ -67,14 +67,14 @@ function counthits(gt::IntegerVector, rklst::IntegerMatrix, ks::IntegerVector)
 end
 
 
-hitrate(gt::IntegerVector, rklst::IntegerMatrix, k::Integer) = 
+hitrate(gt::IntegerVector, rklst::IntegerMatrix, k::Integer) =
     (counthits(gt, rklst, k) / length(gt))::Float64
 
 function hitrates(gt::IntegerVector, rklst::IntegerMatrix, ks::IntegerVector)
     n = length(gt)
     h = counthits(gt, rklst, ks)
     nk = length(ks)
-    r = Array{Float64}(nk)
+    r = Array{Float64}(undef, nk)
     for i = 1:nk
         r[i] = h[i] / n
     end
@@ -178,7 +178,7 @@ length(v::BinaryThresPredVec) = length(v.scores)
 getindex(v::BinaryThresPredVec, i::Integer) = !lt(v.ord, v.scores[i], v.thres)
 
 # compute roc numbers based on scores & threshold
-roc(gt::IntegerVector, scores::RealVector, t::Real, ord::Ordering) = 
+roc(gt::IntegerVector, scores::RealVector, t::Real, ord::Ordering) =
     _roc(gt, BinaryThresPredVec(scores, t, ord))
 
 roc(gt::IntegerVector, scores::RealVector, thres::Real) =
@@ -211,7 +211,7 @@ length(v::ThresPredVec) = length(v.preds)
 getindex(v::ThresPredVec, i::Integer) = ifelse(lt(v.ord, v.scores[i], v.thres), 0, v.preds[i])
 
 # compute roc numbers based on predictions & scores & threshold
-roc(gt::IntegerVector, preds::Tuple{PV,SV}, t::Real, ord::Ordering) where {PV<:IntegerVector,SV<:RealVector} = 
+roc(gt::IntegerVector, preds::Tuple{PV,SV}, t::Real, ord::Ordering) where {PV<:IntegerVector,SV<:RealVector} =
     _roc(gt, ThresPredVec(preds..., t, ord))
 
 roc(gt::IntegerVector, preds::Tuple{PV,SV}, thres::Real) where {PV<:IntegerVector,SV<:RealVector} =
@@ -246,10 +246,10 @@ end
 
 find_thresbin(x::Real, thresholds::RealVector) = find_thresbin(x, thresholds, Forward)
 
-lin_thresholds(scores::RealArray, n::Integer, ord::ForwardOrdering) = 
+lin_thresholds(scores::RealArray, n::Integer, ord::ForwardOrdering) =
     ((s0, s1) = extrema(scores); intv = (s1 - s0) / (n-1); s0:intv:s1)
 
-lin_thresholds(scores::RealArray, n::Integer, ord::ReverseOrdering{ForwardOrdering}) = 
+lin_thresholds(scores::RealArray, n::Integer, ord::ReverseOrdering{ForwardOrdering}) =
     ((s0, s1) = extrema(scores); intv = (s0 - s1) / (n-1); s1:intv:s0)
 
 # roc for binary predictions
@@ -278,7 +278,7 @@ function roc(gt::IntegerVector, scores::RealVector, thresholds::RealVector, ord:
     end
 
     # produce results
-    r = Array{ROCNums{Int}}(nt)
+    r = Array{ROCNums{Int}}(undef, nt)
     fn = 0
     tn = 0
     @inbounds for i = 1:nt
@@ -293,7 +293,7 @@ end
 
 roc(gt::IntegerVector, scores::RealVector, thresholds::RealVector) = roc(gt, scores, thresholds, Forward)
 
-roc(gt::IntegerVector, scores::RealVector, n::Integer, ord::Ordering) = 
+roc(gt::IntegerVector, scores::RealVector, n::Integer, ord::Ordering) =
     roc(gt, scores, lin_thresholds(scores, n, ord), ord)
 
 roc(gt::IntegerVector, scores::RealVector, n::Integer) = roc(gt, scores, n, Forward)
@@ -341,7 +341,7 @@ function roc(
     end
 
     # produce results
-    r = Array{ROCNums{Int}}(nt)
+    r = Array{ROCNums{Int}}(undef, nt)
     fn = 0
     tn = 0
     @inbounds for i = 1:nt
@@ -357,15 +357,14 @@ end
 roc(gt::IntegerVector, preds::Tuple{PV,SV}, thresholds::RealVector) where {PV<:IntegerVector, SV<:RealVector} =
     roc(gt, preds, thresholds, Forward)
 
-roc(gt::IntegerVector, preds::Tuple{PV,SV}, n::Integer, ord::Ordering) where {PV<:IntegerVector, SV<:RealVector} = 
+roc(gt::IntegerVector, preds::Tuple{PV,SV}, n::Integer, ord::Ordering) where {PV<:IntegerVector, SV<:RealVector} =
     roc(gt, preds, lin_thresholds(preds[2],n,ord), ord)
 
-roc(gt::IntegerVector, preds::Tuple{PV,SV}, n::Integer) where {PV<:IntegerVector, SV<:RealVector} = 
+roc(gt::IntegerVector, preds::Tuple{PV,SV}, n::Integer) where {PV<:IntegerVector, SV<:RealVector} =
     roc(gt, preds, n, Forward)
 
-roc(gt::IntegerVector, preds::Tuple{PV,SV}, ord::Ordering) where {PV<:IntegerVector, SV<:RealVector} = 
+roc(gt::IntegerVector, preds::Tuple{PV,SV}, ord::Ordering) where {PV<:IntegerVector, SV<:RealVector} =
     roc(gt, preds, 100, ord)
 
-roc(gt::IntegerVector, preds::Tuple{PV,SV}) where {PV<:IntegerVector, SV<:RealVector} = 
+roc(gt::IntegerVector, preds::Tuple{PV,SV}) where {PV<:IntegerVector, SV<:RealVector} =
     roc(gt, preds, Forward)
-