MLJ table (#262)

* integrate MLJ wrapper with Tables API

* integrate MLJ wrapper with Tables API

* adapt MLJ wrapper to Tables API

* fix support for rowtables

* mix for rowtables input

* replace @Spawn with @threads

* MLJ test

GPU launcher fixes

* up benchmarks

* add back Matrix support in MLJ

Project.toml

@@ -1,7 +1,7 @@
 name = "EvoTrees"
 uuid = "f6006082-12f8-11e9-0c9c-0d5d367ab1e5"
 authors = ["jeremiedb <[email protected]>"]
-version = "0.16.4"
+version = "0.16.5"
 
 [deps]
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"

benchmarks/regressor-df.jl

@@ -6,7 +6,7 @@ using EvoTrees
 using DataFrames
 using BenchmarkTools
 using Random: seed!
-import CUDA
+# import CUDA
 
 nobs = Int(1e6)
 num_feat = Int(100)
@@ -45,13 +45,13 @@ params_xgb = Dict(
     :max_bin => 64,
 )
 
-# dtrain = DMatrix(x_train, y_train)
-# watchlist = Dict("train" => DMatrix(x_train, y_train))
-# @time m_xgb = xgboost(dtrain; watchlist, nthread=nthread, verbosity=0, eval_metric=metric_xgb, params_xgb...);
-# # @btime m_xgb = xgboost($dtrain; watchlist, nthread=nthread, verbosity=0, eval_metric=metric_xgb, params_xgb...);
-# @info "xgboost predict:"
-# @time pred_xgb = XGBoost.predict(m_xgb, x_train);
-# # @btime XGBoost.predict($m_xgb, $x_train);
+dtrain = DMatrix(x_train, y_train)
+watchlist = Dict("train" => DMatrix(x_train, y_train))
+@time m_xgb = xgboost(dtrain; watchlist, nthread=nthread, verbosity=0, eval_metric=metric_xgb, params_xgb...);
+# @btime m_xgb = xgboost($dtrain; watchlist, nthread=nthread, verbosity=0, eval_metric=metric_xgb, params_xgb...);
+@info "xgboost predict:"
+@time pred_xgb = XGBoost.predict(m_xgb, x_train);
+# @btime XGBoost.predict($m_xgb, $x_train);
 
 @info "EvoTrees"
 dtrain = DataFrame(x_train, :auto)

benchmarks/regressor-mlj.jl

@@ -0,0 +1,116 @@
+using Revise
+using Statistics
+using StatsBase: sample
+using EvoTrees
+using DataFrames
+using BenchmarkTools
+using Random: seed!
+import CUDA
+using MLJ
+
+nobs = Int(2e6)
+num_feat = Int(100)
+nrounds = 200
+T = Float64
+nthread = Base.Threads.nthreads()
+@info "testing with: $nobs observations | $num_feat features. nthread: $nthread"
+seed!(123)
+x_train = rand(T, nobs, num_feat)
+y_train = rand(T, size(x_train, 1))
+
+@info nthread
+loss = "mse"
+if loss == "mse"
+    loss_evo = :mse
+    metric_evo = :mae
+elseif loss == "logloss"
+    loss_evo = :logloss
+    metric_evo = :logloss
+end
+
+@info "EvoTrees"
+dtrain = DataFrame(x_train, :auto)
+# dtrain.y .= y_train
+# target_name = "y"
+verbosity = 0
+
+params_evo = EvoTreeRegressor(
+    loss=loss_evo,
+    nrounds=nrounds,
+    alpha=0.5,
+    lambda=0.0,
+    gamma=0.0,
+    eta=0.05,
+    max_depth=6,
+    min_weight=1.0,
+    rowsample=0.5,
+    colsample=0.5,
+    nbins=64,
+    rng=123,
+)
+
+@info "EvoTrees CPU"
+device = "cpu"
+
+iterated_model = IteratedModel(
+    model=params_evo,
+    resampling=Holdout(; fraction_train=0.5),
+    measures=rmse,
+    controls=[Step(5),
+        Patience(200),
+        NumberLimit(40)],
+    retrain=false)
+
+mach = machine(iterated_model, dtrain, y_train)
+@time fit!(mach);
+
+@info "init"
+@time m_df, cache_df = EvoTrees.init(params_evo, dtrain; target_name);
+
+# @info "train - no eval"
+# @time m_evo_df = fit_evotree(params_evo, dtrain; target_name, device, verbosity, print_every_n=100);
+# @time m_evo_df = fit_evotree(params_evo, dtrain; target_name, device, verbosity, print_every_n=100);
+
+@info "train - eval"
+@time m_evo = fit_evotree(params_evo, dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+@time m_evo = fit_evotree(params_evo, dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+# @time m_evo = fit_evotree(params_evo, dtrain; target_name, device);
+# @btime fit_evotree($params_evo, $dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+@info "predict"
+@time pred_evo = m_evo(dtrain);
+@btime m_evo($dtrain);
+
+@info "EvoTrees GPU"
+device = "gpu"
+@info "train"
+@time m_evo = fit_evotree(params_evo, dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+@time m_evo = fit_evotree(params_evo, dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+# @btime m_evo = fit_evotree($params_evo, $dtrain; target_name, device);
+# @btime fit_evotree($params_evo, $dtrain; target_name, deval=dtrain, metric=metric_evo, device, verbosity, print_every_n=100);
+@info "predict"
+@time pred_evo = m_evo(dtrain; device);
+@btime m_evo($dtrain; device);
+
+
+using MLJBase
+using MLJModels
+using Tables
+
+EvoTreeBooster = @load EvoTreeRegressor
+booster = EvoTreeBooster()
+
+X, y = make_regression(1000, 5)
+
+# this works:
+mach = machine(booster, X, y) |> fit!
+
+# this doesn't
+X, y = make_regression(1_000_000, 100);
+@time X = DataFrame(X);
+@time X = Tables.rowtable(X);
+@time X = Tables.columntable(X);
+
+mach = machine(booster, X, y) |> fit!
+
+schema = Tables.schema(dtrain)
+schema.names

benchmarks/regressor.jl

@@ -6,15 +6,15 @@ using XGBoost
 using EvoTrees
 using BenchmarkTools
 using Random: seed!
-import CUDA
+# import CUDA
 
 ### v.0.15.1
 # desktop | 1e6 | depth 11 | cpu: 37.2s
 # desktop | 10e6 | depth 11 | cpu
 
-### perf depth
-# desktop | 1e6 | depth 11 | cpu: 28s gpu: 73 sec  | xgboost: 26s
-# desktop | 10e6 | depth 11 | cpu 205s gpu: 109 sec | xgboost 260s
+### v0.16.5
+# desktop | 1e6 | depth 11 | cpu: 31s gpu: 50 sec  | xgboost cpu: 26s
+# desktop | 10e6 | depth 11 | cpu 200s gpu: 80 sec | xgboost cpu: 267s
 
 #threads
 # laptop depth 6: 12.717845 seconds (2.08 M allocations: 466.228 MiB)

experiments/MLJ.jl

@@ -68,7 +68,6 @@ config = EvoTreeClassifier(
     gamma = 0.0,
     nbins = 32,
     nrounds = 200,
-    device = "cpu"
 )
 model = fit_evotree(config; x_train, y_train);
 model = fit_evotree(config; x_train, y_train, x_eval = x_train, y_eval = y_train, metric=:mlogloss, print_every_n=10, early_stopping_rounds=25);