Merge pull request #176 from Evovest/dev

fix GC GPU & docs typo

Project.toml

@@ -1,7 +1,7 @@
 authors = ["jeremiedb <[email protected]>"]
 name = "EvoTrees"
 uuid = "f6006082-12f8-11e9-0c9c-0d5d367ab1e5"
-version = "0.12.3"
+version = "0.12.4"
 
 [deps]
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"

docs/src/index.md

@@ -1,4 +1,4 @@
-# [EvoTress.jl](https://github.com/Evovest/EvoTrees.jl)
+# [EvoTrees.jl](https://github.com/Evovest/EvoTrees.jl)
 
 A Julia implementation of boosted trees with CPU and GPU support. Efficient histogram based algorithms with support for multiple loss functions, including various regressions, multi-classification and Gaussian max likelihood. 
 

experiments/benchmarks_v2-gpu.jl

@@ -0,0 +1,41 @@
+using Revise
+using Statistics
+using StatsBase: sample
+using EvoTrees
+using BenchmarkTools
+using CUDA
+
+nrounds = 200
+nthread = Base.Threads.nthreads()
+
+@info nthread
+
+# EvoTrees params
+params_evo = EvoTreeRegressor(
+    T=Float32,
+    loss="linear",
+    nrounds=nrounds,
+    alpha=0.5,
+    lambda=0.0,
+    gamma=0.0,
+    eta=0.05,
+    max_depth=6,
+    min_weight=1.0,
+    rowsample=1.0,
+    colsample=1.0,
+    nbins=64,
+    device = "gpu"
+)
+
+nobs = Int(11_664_400)
+num_feat = Int(36)
+@info "testing with: $nobs observations | $num_feat features."
+x_train = rand(nobs, num_feat)
+y_train = rand(size(x_train, 1))
+
+@info "evotrees train GPU:"
+@time m_evo_gpu = fit_evotree(params_evo; x_train, y_train);
+for i in 1:5
+    @time m_evo_gpu = fit_evotree(params_evo; x_train, y_train);
+end
+# @time m_evo = fit_evotree(params_evo; x_train, y_train, x_eval=x_train, y_eval=y_train, metric=metric_evo, print_every_n=100);

src/fit.jl

@@ -393,6 +393,10 @@ function fit_evotree(
         model.metric.iter = metric_best.iter
         model.metric.metric = metric_best.metric
     end
+    if params.device == "gpu"
+        GC.gc(true)
+        CUDA.reclaim()
+    end
     params.nrounds = nrounds_max
     return model
 end

src/gpu/fit_gpu.jl

@@ -135,7 +135,6 @@ function grow_evotree!(evotree::GBTreeGPU{L,T,S}, cache) where {L,T,S}
         predict!(cache.pred, tree, cache.x, cache.K)
     end # end of nrounds
     cache.params.nrounds = params.nrounds
-    CUDA.reclaim()
     return evotree
 end