Jdb/cat compat (#315)

* categorical init

* test

Author: jeremiedb

Project.toml

@@ -1,7 +1,7 @@
 name = "EvoTrees"
 uuid = "f6006082-12f8-11e9-0c9c-0d5d367ab1e5"
 authors = ["jeremiedb <[email protected]>"]
-version = "0.18.1"
+version = "0.18.2"
 
 [deps]
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"
@@ -30,7 +30,7 @@ CUDA = "3.0, 4.0, 5.0"
 CategoricalArrays = "1"
 Distributions = "0.24, 0.25"
 KernelAbstractions = "0.9"
-MLJModelInterface = "0.3, 0.4, 1.0"
+MLJModelInterface = "1.2.1"
 NetworkLayout = "0.4"
 Random = "1"
 RecipesBase = "1"

benchmarks/softmax.jl

@@ -1,8 +1,8 @@
-using Revise
 using Statistics
 using StatsBase: sample
 using XGBoost
 using EvoTrees
+using EvoTrees: fit
 using BenchmarkTools
 import CUDA
 
@@ -54,22 +54,23 @@ params_evo = EvoTreeClassifier(;
     min_weight=1.0,
     rowsample=0.5,
     colsample=0.5,
-    nbins=64)
+    nbins=64
+)
 
 @info "EvoTrees CPU"
 params_evo.device = :cpu
 @info "train - eval"
-@time m_evo = fit_evotree(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
-@time m_evo = fit_evotree(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
+@time m_evo = fit(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
+@time m_evo = fit(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
 @info "evotrees predict CPU:"
 @time pred_evo = m_evo(x_train);
 @btime m_evo($x_train);
 
 @info "evotrees train GPU:"
 params_evo.device = :gpu
 @info "train - eval"
-@time m_evo = fit_evotree(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
-@time m_evo = fit_evotree(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
+@time m_evo = fit(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
+@time m_evo = fit(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, print_every_n=100);
 # @btime fit_evotree($params_evo; x_train=$x_train, y_train=$y_train, x_eval=$x_train, y_eval=$y_train, metric=metric_evo);
 @info "evotrees predict GPU:"
 @time pred_evo = m_evo(x_train; device=:gpu);

ext/EvoTreesCUDAExt/init.jl

@@ -27,8 +27,8 @@ function EvoTrees.init_core(params::EvoTrees.EvoTypes, ::Type{<:EvoTrees.GPU}, d
             target_isordered = EvoTrees.isordered(y_train)
             y = UInt32.(EvoTrees.CategoricalArrays.levelcode.(y_train))
         elseif eltype(y_train) <: Integer || eltype(y_train) <: Bool || eltype(y_train) <: String || eltype(y_train) <: Char
-            target_levels = sort(unique(y_train))
-            yc = EvoTrees.CategoricalVector(y_train, levels=target_levels)
+            yc = EvoTrees.CategoricalArrays.categorical(y_train, levels=sort(unique(y_train)), ordered=false)
+            target_levels = EvoTrees.CategoricalArrays.levels(yc)
             y = UInt32.(EvoTrees.CategoricalArrays.levelcode.(yc))
         else
             @error "Invalid target eltype: $(eltype(y_train))"

src/init.jl

@@ -30,8 +30,8 @@ function init_core(params::EvoTypes, ::Type{CPU}, data, feature_names, y_train,
             target_isordered = isordered(y_train)
             y = UInt32.(CategoricalArrays.levelcode.(y_train))
         elseif eltype(y_train) <: Integer || eltype(y_train) <: Bool || eltype(y_train) <: String || eltype(y_train) <: Char
-            target_levels = sort(unique(y_train))
-            yc = CategoricalVector(y_train, levels=target_levels)
+            yc = categorical(y_train, levels=sort(unique(y_train)), ordered=false)
+            target_levels = CategoricalArrays.levels(yc)
             y = UInt32.(CategoricalArrays.levelcode.(yc))
         else
             @error "Invalid target eltype: $(eltype(y_train))"