fix hist cumul

sethaxen · Dec 25, 2022 · 0e30320 · 0e30320
1 parent 2d7f99c
commit 0e30320
Show file tree

Hide file tree

Showing 6 changed files with 8 additions and 85 deletions.
diff --git a/experiments/debug-softmax-split.jl b/experiments/debug-softmax-split.jl
@@ -21,7 +21,7 @@ Xtrain, ytrain = MLJBase.reformat(model, selectrows(X, train), selectrows(y, tra
 rng = StableRNG(6)
 params_evo = EvoTreeClassifier(;
     T = Float32,
-    nrounds = 5,
+    nrounds = 200,
     lambda = 0.0,
     gamma = 0.0,
     eta = 0.1,

diff --git a/experiments/readme_plots_gpu.jl b/experiments/readme_plots_gpu.jl
@@ -30,9 +30,9 @@ y_train, y_eval = Y[i_train], Y[i_eval]
 # linear
 params1 = EvoTreeRegressor(T=Float32,
     loss=:linear,
-    nrounds=1, nbins=16,
+    nrounds=500, nbins=64,
     lambda=0.1, gamma=0.1, eta=0.1,
-    max_depth=3, min_weight=1.0,
+    max_depth=6, min_weight=1.0,
     rowsample=0.5, colsample=1.0,
     device="gpu")
 
@@ -140,4 +140,4 @@ plot!(x_train[:, 1][x_perm], pred_train_gaussian[x_perm, 1], color="navy", linew
 plot!(x_train[:, 1][x_perm], pred_train_gaussian[x_perm, 2], color="darkred", linewidth=1.5, label="sigma")
 plot!(x_train[:, 1][x_perm], pred_q20[x_perm, 1], color="green", linewidth=1.5, label="q20")
 plot!(x_train[:, 1][x_perm], pred_q80[x_perm, 1], color="green", linewidth=1.5, label="q80")
-savefig("figures/gaussian-sinus-gpu.png")
+savefig("figures/gaussian-sinus-gpu.png")
diff --git a/figures/gaussian-sinus-gpu.png b/figures/gaussian-sinus-gpu.png
diff --git a/figures/regression_sinus_gpu.png b/figures/regression_sinus_gpu.png
diff --git a/src/find_split.jl b/src/find_split.jl
@@ -237,85 +237,20 @@ end
 
 Generic fallback
 """
-# function update_gains!(
-#     node::TrainNode,
-#     js::Vector,
-#     params::EvoTypes{L,T},
-#     K,
-#     monotone_constraints,
-# ) where {L,T}
-
-#     h = node.h
-#     hL = node.hL
-#     hR = node.hR
-#     gains = node.gains
-
-#     cumsum!(hL, h, dims = 2)
-#     hR .= view(hL, :, params.nbins:params.nbins, :) .- hL
-#     @info "minimum(hR[3,:,:])" minimum(hR[3,:,:])
-
-#     @inbounds for j in js
-#         monotone_constraint = monotone_constraints[j]
-#         @inbounds for bin = 1:params.nbins
-#             if hL[end, bin, j] > params.min_weight && hR[end, bin, j] > params.min_weight
-#                 if monotone_constraint != 0
-#                     predL = pred_scalar(view(hL, :, bin, j), params)
-#                     predR = pred_scalar(view(hR, :, bin, j), params)
-#                 end
-#                 if (monotone_constraint == 0) ||
-#                    (monotone_constraint == -1 && predL > predR) ||
-#                    (monotone_constraint == 1 && predL < predR)
-
-#                     gains[bin, j] =
-#                         get_gain(params, view(hL, :, bin, j)) +
-#                         get_gain(params, view(hR, :, bin, j))
-#                 end
-#             end
-#         end
-#     end
-#     return nothing
-# end
-
 function update_gains!(
     node::TrainNode,
     js::Vector,
     params::EvoTypes{L,T},
-    K,
     monotone_constraints,
 ) where {L,T}
 
-    KK = 2 * K + 1
     h = node.h
     hL = node.hL
     hR = node.hR
     gains = node.gains
-    ∑ = node.∑
-
-    @inbounds for j in js
-        @inbounds for k = 1:KK
-            val = h[k, 1, j]
-            hL[k, 1, j] = val
-            hR[k, 1, j] = ∑[k] - val
-        end
-        @inbounds for bin = 2:params.nbins
-            @inbounds for k = 1:KK
-                val = h[k, bin, j]
-                hL[k, bin, j] = hL[k, bin-1, j] + val
-                hR[k, bin, j] = hR[k, bin-1, j] - val
-            end
-        end
-    end
 
-    # hL2 = copy(hL) .* 0
-    # hR2 = copy(hR) .* 0
-    # cumsum!(hL2, h, dims = 2)
-    # hR2 .= view(hL2, :, params.nbins:params.nbins, :) .- hL2
-    # @info "max abs diff hL" maximum(abs.(hL[3, :, :] .- hL2[3, :, :]))
-    # @info "max abs diff hR" maximum(abs.(hR[3, :, :] .- hR2[3, :, :]))
-    @info "minimum(hR[3,:,:])" minimum(hR[3, :, :])
-    # @info "minimum(hR2[3,:,:])" minimum(hR2[3, :, :])
-    # @info "node.∑" node.∑
-    # @info "sum(node.h, dims=2)" sum(node.h, dims=2)
+    cumsum!(hL, h, dims = 2)
+    hR .= view(hL, :, params.nbins:params.nbins, :) .- hL
 
     @inbounds for j in js
         monotone_constraint = monotone_constraints[j]

diff --git a/src/fit.jl b/src/fit.jl
@@ -192,7 +192,7 @@ function grow_tree!(
     # grow while there are remaining active nodes
     while length(n_current) > 0 && depth <= params.max_depth
         offset = 0 # identifies breakpoint for each node set within a depth
-
+        
         if depth < params.max_depth
             for n_id in eachindex(n_current)
                 n = n_current[n_id]
@@ -212,20 +212,8 @@ function grow_tree!(
             if depth == params.max_depth || nodes[n].∑[end] <= params.min_weight
                 pred_leaf_cpu!(tree.pred, n, nodes[n].∑, params, ∇, nodes[n].is)
             else
-                update_gains!(nodes[n], js, params, K, monotone_constraints)
+                update_gains!(nodes[n], js, params, monotone_constraints)
                 best = findmax(nodes[n].gains)
-                # if depth in [2]
-                #     @info "minimum(nodes[n].h[3,:,:])" minimum(nodes[n].h[3,:,:])
-                #     @info "minimum(nodes[n].h[4,:,:])" minimum(nodes[n].h[4,:,:])
-                #     @info "nodes[n].hL" nodes[n].hL[:, best[2][1], best[2][2]]
-                #     @info "nodes[n].hR" nodes[n].hR[:, best[2][1], best[2][2]]
-                # end
-                if depth in [2,3,4]
-                    @info "depth" depth
-                    # @info "best" best
-                    # @info "nodes[n].gain" nodes[n].gain
-                    # @info "nodes[n].∑" nodes[n].∑
-                end
                 if best[2][1] != params.nbins && best[1] > nodes[n].gain + params.gamma
                     tree.gain[n] = best[1] - nodes[n].gain
                     tree.cond_bin[n] = best[2][1]