Skip to content

Commit

Permalink
Merge e3812c9 into ba77a3c
Browse files Browse the repository at this point in the history
  • Loading branch information
@benevolent0505
benevolent0505 committed Nov 27, 2017
2 parents ba77a3c + e3812c9 commit 05b86e5
Show file tree
Hide file tree
Showing 2 changed files with 20 additions and 14 deletions.
18 changes: 12 additions & 6 deletions src/ScikitLearnAPI.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,17 +4,17 @@ import ScikitLearnBase: predict, predict_proba,
set_params!

function predict(model::Union{AbstractSVC, AbstractSVR} , X::AbstractArray)
(p,d) = svmpredict(model.fit, X)
(p,d) = svmpredict(model.fit, X')
return p
end

function predict(model::LinearSVC, X::AbstractArray)
(p,d) = LIBLINEAR.linear_predict(model.fit, X)
(p,d) = LIBLINEAR.linear_predict(model.fit, X')
return p
end

function transform(model::OneClassSVM, X::AbstractArray)
(p,d) = svmpredict(model.fit, X)
(p,d) = svmpredict(model.fit, X')
return p
end

Expand All @@ -25,43 +25,49 @@ SVC(;kernel::Kernel.KERNEL = Kernel.RadialBasis, gamma::Union{Float64,Symbol} =
verbose::Bool = false) = SVC(kernel, gamma,
weights, cost, degree, coef0, tolerance, shrinking,
probability, verbose, nothing)
@declare_hyperparameters(SVC, [:kernel, :gamma, :weights, :cost, :degree, :coef0, :tolerance])

NuSVC(;kernel::Kernel.KERNEL = Kernel.RadialBasis, gamma::Union{Float64,Symbol} = :auto,
weights = nothing, nu::Float64 = 0.5, cost::Float64 = 1.0,
degree::Int32 = Int32(3), coef0::Float64 = 0.,
tolerance::Float64 = .001, shrinking::Bool = true,
verbose::Bool = false,) = NuSVC(kernel, gamma, weights, nu, cost,
degree, coef0, tolerance, shrinking, verbose, nothing)
@declare_hyperparameters(NuSVC, [:kernel, :gamma, :weights, :nu, :cost, :degree, :coef0, :tolerance])

OneClassSVM(;kernel::Kernel.KERNEL = Kernel.RadialBasis, gamma::Union{Float64,Symbol} = :auto,
nu::Float64 = 0.1, cost::Float64 = 1.0, degree::Int32 = Int32(3),
coef0::Float64 = 0.0, tolerance::Float64 = .001,
shrinking::Bool = true,
verbose::Bool = false,) = OneClassSVM(kernel, gamma, nu, cost,
degree, coef0, tolerance, shrinking, verbose, nothing)
@declare_hyperparameters(OneClassSVM, [:kernel, :gamma, :nu, :cost, :degree, :coef0, :tolerance])

NuSVR(;kernel::Kernel.KERNEL = Kernel.RadialBasis, gamma::Union{Float64,Symbol} = :auto,
nu::Float64 = 0.5, cost::Float64 = 1.0, degree::Int32 = Int32(3), coef0::Float64 = 0.,
tolerance::Float64 = .001, shrinking::Bool = true,
verbose::Bool = false,) = NuSVR(kernel, gamma, nu, cost,
degree, coef0, tolerance, shrinking, verbose, nothing)
@declare_hyperparameters(NuSVR, [:kernel, :gamma, :nu, :cost, :degree, :coef0, :tolerance])

EpsilonSVR(;kernel::Kernel.KERNEL = Kernel.RadialBasis, gamma::Union{Float64,Symbol} = :auto,
epsilon::Float64 = 0.1, cost::Float64 = 1.0,
degree::Int32 = Int32(3), coef0::Float64 = 0.,
tolerance::Float64 = .001, shrinking::Bool = true,
verbose::Bool = false,) = EpsilonSVR(kernel, gamma, epsilon, cost,
degree, coef0, tolerance, shrinking, verbose, nothing)
@declare_hyperparameters(EpsilonSVR, [:kernel, :gamma, :epsilon, :cost, :degree, :coef0, :tolerance])

LinearSVC(;solver = Linearsolver.L2R_L2LOSS_SVC_DUAL,
weights::Union{Dict, Void} = nothing, tolerance::Float64=Inf,
cost::Float64 = 1.0, p::Float64 = 0.1, bias::Float64 = -1.0,
verbose::Bool = false) = LinearSVC(solver, weights, tolerance,
cost, p, bias, verbose, nothing)
@declare_hyperparameters(LinearSVC, [:solver, :weights, :tolerance, :cost, :p, :bias])

function fit!(model::Union{AbstractSVC,AbstractSVR}, X::AbstractMatrix, y::Vector=[])
#Build arguments for calling svmtrain
model.gamma == :auto && (model.gamma = 1.0/size(X, 1))
model.gamma == :auto && (model.gamma = 1.0/size(X', 1))
kwargs = Tuple{Symbol, Any}[]
push!(kwargs, (:svmtype, typeof(model)))
for fn in fieldnames(model)
Expand All @@ -70,7 +76,7 @@ function fit!(model::Union{AbstractSVC,AbstractSVR}, X::AbstractMatrix, y::Vecto
end
end

model.fit = svmtrain(X, y; kwargs...)
model.fit = svmtrain(X', y; kwargs...)
return(model)
end

Expand All @@ -92,7 +98,7 @@ function get_params(model::Union{AbstractSVC,AbstractSVR, LinearSVC})
end

function fit!(model::LinearSVC, X::AbstractMatrix, y::Vector)
model.fit = LIBLINEAR.linear_train(y, X, solver_type = Int32(model.solver),
model.fit = LIBLINEAR.linear_train(y, X', solver_type = Int32(model.solver),
weights = model.weights, C = model.cost, bias = model.bias,
p = model.p, eps = model.tolerance, verbose = model.verbose)
return(model)
Expand Down
16 changes: 8 additions & 8 deletions test/runtests.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,8 +10,8 @@ gc()
correct = Bool[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1]
@test (class .== labels[2:2:end]) == correct

skmodel = fit!(SVC(), instances[:,1:2:end], labels[1:2:end])
skclass = predict(skmodel, instances[:, 2:2:end])
skmodel = fit!(SVC(), instances[:,1:2:end]', labels[1:2:end])
skclass = predict(skmodel, instances[:, 2:2:end]')
@test skclass == class

model = svmtrain(sparse(instances[:, 1:2:end]), labels[1:2:end]; verbose=true)
Expand All @@ -28,21 +28,21 @@ y = Array(whiteside[:Temp])
m = svmtrain(X, y, svmtype = EpsilonSVR, cost = 10., gamma = 1.)
yeps, d = svmpredict(m, X)
@test_approx_eq(sum(yeps - y), 7.455509045783046)
skm = fit!(EpsilonSVR(cost = 10., gamma = 1.), X, y)
ysk = predict(skm, X)
skm = fit!(EpsilonSVR(cost = 10., gamma = 1.), X', y)
ysk = predict(skm, X')
@test yeps == ysk

nu1 = svmtrain(X, y, svmtype = NuSVR, cost = 10.,
nu = .7, gamma = 2., tolerance = .001)
ynu1, d = svmpredict(nu1, X)
@test_approx_eq(sum(ynu1 - y), 14.184665717092)
sknu1 = fit!(NuSVR(cost = 10., nu=.7, gamma = 2.), X, y)
ysknu1 = predict(sknu1, X)
sknu1 = fit!(NuSVR(cost = 10., nu=.7, gamma = 2.), X', y)
ysknu1 = predict(sknu1, X')
@test ysknu1 == ynu1

nu2 = svmtrain(X, y, svmtype = NuSVR, cost = 10., nu = .9)
ynu2, d =svmpredict(nu2, X)
@test_approx_eq(sum(ynu2 - y), 6.686819661799177)
sknu2 = fit!(NuSVR(cost = 10., nu=.9), X, y)
ysknu2 = predict(sknu2, X)
sknu2 = fit!(NuSVR(cost = 10., nu=.9), X', y)
ysknu2 = predict(sknu2, X')
@test ysknu2 == ynu2

0 comments on commit 05b86e5

Please sign in to comment.