preparing merge

aclai-lab · Jan 20, 2025 · 622a875 · 622a875
1 parent 0567d15
commit 622a875
Showing 1 changed file with 27 additions and 48 deletions.
diff --git a/src/apply.jl b/src/apply.jl
@@ -16,9 +16,8 @@ function apply end
 apply_model = apply
 
 # apply_tree = apply_model
-@deprecate apply_tree apply_model
-# apply_forest = apply_model
-@deprecate apply_forest apply_model
+# @deprecate apply_tree apply_model
+# @deprecate apply_forest apply_model
 
 function apply_proba end
 
@@ -351,72 +350,52 @@ end
 
 # use an array of trees to test features
 function sprinkle(
-    trees::AbstractVector{<:DTree},
+    trees::AbstractVector{<:DTree{<:L}},
     Xs,
     Y::AbstractVector{<:L};
     print_progress = !(Xs isa MultiLogiset),
     tree_weights::Union{AbstractMatrix{Z},AbstractVector{Z},Nothing} = nothing,
     suppress_parity_warning = false,
-) where {L<:Label, Z<:Real}
+) where {L<:Label,Z<:Real}
     @logmsg LogDetail "sprinkle..."
-
     trees = deepcopy(trees)
     ntrees = length(trees)
     _ninstances = ninstances(Xs)
 
-    # if !(tree_weights isa AbstractMatrix)
-    #     if isnothing(tree_weights)
-    #         tree_weights = Ones{Int}(length(trees), ninstances(Xs)) # TODO optimize?
-    #     elseif tree_weights isa AbstractVector
-    #         tree_weights = hcat([tree_weights for i_instance in 1:ninstances(Xs)]...)
-    #     else
-    #         @show typeof(tree_weights)
-    #         error("Unexpected tree_weights encountered $(tree_weights).")
-    #     end
-    # end
-
-    if isnothing(tree_weights)
-        tree_weights = Ones{Int}(ntrees)
+    if !(tree_weights isa AbstractMatrix)
+        if isnothing(tree_weights)
+            tree_weights = Ones{Int}(length(trees), ninstances(Xs)) # TODO optimize?
+        elseif tree_weights isa AbstractVector
+            tree_weights = hcat([tree_weights for i_instance in 1:ninstances(Xs)]...)
+        else
+            @show typeof(tree_weights)
+            error("Unexpected tree_weights encountered $(tree_weights).")
+        end
     end
 
-    # @assert length(trees) == size(tree_weights, 1) "Each label must have a corresponding weight: labels length is $(length(labels)) and weights length is $(length(weights))."
-    # @assert ninstances(Xs) == size(tree_weights, 2) "Each label must have a corresponding weight: labels length is $(length(labels)) and weights length is $(length(weights))."
+    @assert length(trees) == size(tree_weights, 1) "Each label must have a corresponding weight: labels length is $(length(labels)) and weights length is $(length(weights))."
+    @assert ninstances(Xs) == size(tree_weights, 2) "Each label must have a corresponding weight: labels length is $(length(labels)) and weights length is $(length(weights))."
 
-    predictions = Vector{L}(undef, _ninstances)
+    # apply each tree to the whole dataset
     _predictions = Matrix{L}(undef, ntrees, _ninstances)
-
     if print_progress
         p = Progress(ntrees; dt = 1, desc = "Applying trees...")
     end
-
-    # apply each tree to the whole dataset
     Threads.@threads for i_tree in 1:ntrees
         _predictions[i_tree,:], trees[i_tree] = sprinkle(trees[i_tree], Xs, Y; print_progress = false)
+        print_progress && next!(p)
     end
 
     # for each instance, aggregate the predictions
-    for i_instance in 1:_ninstances
-        _counts = Dict{L, Float64}()
-
-        Threads.@threads for i_tree in 1:ntrees
-            prediction = _predictions[i_tree, i_instance]
-            _counts[prediction] = get(_counts, prediction, 0.0) + tree_weights[i_tree]
-        end
-
-        top_prediction = trees[1].root.left.prediction
-        top_count = -Inf
-
-        for (k, v) in _counts
-            if v > top_count
-                top_prediction = k
-                top_count = v
-            end
-        end
-        predictions[i_instance] = top_prediction
-
-        print_progress && next!(p)
+    predictions = Vector{L}(undef, _ninstances)
+    Threads.@threads for i_instance in 1:_ninstances
+        predictions[i_instance] = bestguess(
+            _predictions[:,i_instance],
+            tree_weights[:,i_instance];
+            suppress_parity_warning = suppress_parity_warning
+        )
     end
-    # @show predictions
+
     predictions, trees
 end
 
@@ -430,7 +409,7 @@ function sprinkle(
 ) where {L<:Label}
     predictions, trees = begin
         if weight_trees_by == false
-            sprinkle(trees(forest), Xs, Y; kwargs...)
+            sprinkle(ModalDecisionTrees.trees(forest), Xs, Y; kwargs...)
         elseif isa(weight_trees_by, AbstractVector)
             sprinkle(trees(forest), Xs, Y; tree_weights = weight_trees_by, kwargs...)
         # elseif weight_trees_by == :accuracy
@@ -747,4 +726,4 @@ end
 #     metrics
 # end
 
-# tree_walk_metrics(tree::DTree; kwargs...) = tree_walk_metrics(tree.root; kwargs...)
+# tree_walk_metrics(tree::DTree; kwargs...) = tree_walk_metrics(tree.root; kwargs...)