Merge pull request #21 from AsafManela/master

Fix cross-validation and add tests for it

src/cross_validation.jl

@@ -5,7 +5,7 @@ using MLBase
 # which chooses λt at lowest mean OOS deviance.
 
 function CVmin(oosdevs)
-    cvmeans = mean(oosdevs,dims=2)
+    cvmeans = vec(mean(oosdevs,dims=2))
     segCVmin = argmin(cvmeans)
 end
 

test/cross_validation.jl

@@ -1,56 +1,36 @@
-using MLBase
+using MLBase, Random
 
-datapath = joinpath(dirname(@__FILE__), "..","test","data")
+datapath = joinpath(dirname(@__FILE__), "data")
 
-(family, dist, link) = (("gaussian", Normal(), IdentityLink()), ("binomial", Binomial(), LogitLink()), ("poisson", Poisson(), LogLink()))[1]
-data = readcsvmat(joinpath(datapath,"gamlr.$family.data.csv"))
+data = readcsvmat(joinpath(datapath,"gamlr.gaussian.data.csv"))
 y = data[:,1]
 X = data[:,2:end]
-(n,p) = size(X)
-γ = [0 2 10][1]
-fitname = "gamma$γ"
-# get gamlr params and estimates
-params = readtable(joinpath(datapath,"gamlr.$family.$fitname.params.csv"))
-fittable = readtable(joinpath(datapath,"gamlr.$family.$fitname.fit.csv"))
-gcoefs = convert(Matrix{Float64},readcsv(joinpath(datapath,"gamlr.$family.$fitname.coefs.csv")))
-family = params[1,:fit_family]
-λ = nothing #convert(Vector{Float64},fittable[:fit_lambda]) # should be set to nothing evenatually
-# fit julia version
 offset = fill(0.001,size(y))
 
-@time glp = fit(GammaLassoPath, X, y, dist, link, λ=λ,γ=γ,standardize=true, λminratio=0.001, offset=offset)
-path = glp
+path = fit(LassoPath, X, y; offset=offset)
+β = coef(path; select=:all)
 
-plot(path)
+coefsAICc = coef(path;select=:AICc)
+segminAICc = minAICc(path)
+@test segminAICc == 71
+@test coefsAICc == β[:,segminAICc]
 
-@time coefsAICc = coef(path;select=:AICc)
 Random.seed!(13)
-@time coefsCVmin = coef(path;select=:CVmin)
-Random.seed!(13)
-@time coefsCV1se = coef(path;select=:CV1se,nCVfolds=100)
-# fieldnames(path.m.pp)
-# y == path.m.rr.y
-# offset == path.m.rr.offset
-# path.m.pp.X
-#
-# size(path.m.pp.X)
-# size(convert(path.Xnorm)
-
-# gen = LOOCV(nobs(path))
-# T = eltype(λ)
-# offset=Array(T,0)
-ix=1:length(y)
-
-# plot(path)
-
-Random.seed!(13); gen = Kfold(length(y[ix]),10)
-@time segCVmin = cross_validate_path(path;gen=gen)
+gen = Kfold(length(y),10)
+segCVmin = cross_validate_path(path;gen=gen)
+coefsCVmin = coef(path;select=:CVmin)
+@test segCVmin == 71
+@test coefsCVmin == β[:,segCVmin]
 
-Random.seed!(13); gen = Kfold(length(y[ix]),10)
-@time segCVmin = cross_validate_path(path,X[ix,:],y[ix];offset=offset[ix],gen=gen)
-
-Random.seed!(13); gen = Kfold(length(y[ix]),10)
-@time segCV1se = cross_validate_path(path,X[ix,:],y[ix];select=:CV1se,gen=gen,offset=offset[ix])
+Random.seed!(13)
+gen = Kfold(length(y),10)
+segCVmin = cross_validate_path(path,X,y; gen=gen, offset=offset)
+coefsCVmin = coef(path;select=:CVmin)
+@test segCVmin == 71
+@test coefsCVmin == β[:,segCVmin]
 
-λCVmin = path.λ[segCVmin]
-λCV1se = path.λ[segCV1se]
+Random.seed!(13)
+coefsCV1se = coef(path;select=:CV1se,nCVfolds=20)
+segCV1se = cross_validate_path(path,X,y;select=:CV1se,gen=gen,offset=offset)
+@test segCV1se == 42
+@test coefsCV1se == β[:,segCV1se]

test/runtests.jl

@@ -9,5 +9,6 @@ include("lasso.jl")
 include("gammalasso.jl")
 include("fusedlasso.jl")
 include("trendfiltering.jl")
+include("cross_validation.jl")
 
 end