svmkit 0.6.1 → 0.6.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 1d52bf496a762b096a5f6dbeec278a1cae8079b53d6c91cc13c07dca7a799fde
-  data.tar.gz: e5ca2fed307b82e88dfe816691a4715d62a3187c1cad71421a48bea65037b19c
+  metadata.gz: 9655d7d990f37468c79de9713e55f74c6134b00c1cda9471832097c678cb6ded
+  data.tar.gz: fb294e6256d16272e80c2ade2f5b223dd792b14c7fd39caf1de8a9500a3ab55e
 SHA512:
-  metadata.gz: 620370c119300f3f419550609444eba4aa34561a954e8ec26cf6a0d3522cd32cabf1f6875092de5ab0dd202ebf7b772c1d6d6421cd05d90cfeeeeadea3cd0565
-  data.tar.gz: a0d8b5a7b91c4a8e2ffb4312a8082096a2e4fbd411e37bd36752ac19a66f6b9accf22be15894d6a88421b7296bde90ead33b55604085d139e5faec64b97f0f55
+  metadata.gz: f7386071fe57df51bd8223d4945dc67069769464892bd64b972b5c1ba26cd206b7b67d50e600f34d79a3bff9f19803c0fdae06dd92fdf8f6ef87f1d5e982cf2d
+  data.tar.gz: 473a1233e0109672b80b8bf17933366276b0a81fab3c75699fb1a07f92923d66cc3f064e0a2df36e493bf1554aa4c34d567d60607d1802e10ab949eafe1187d3

data/.rubocop.yml

@@ -42,3 +42,6 @@ Style/FormatStringToken:
 
 Style/NumericLiterals:
   Enabled: false
+
+Layout/EmptyLineAfterGuardClause:
+  Enabled: false

data/HISTORY.md

@@ -1,7 +1,10 @@
+# 0.6.2
+- Refactor decision tree classes for improving performance.
+
 # 0.6.1
 - Add abstract class for linear estimator with stochastic gradient descent.
 - Refactor linear estimators to use linear esitmator abstract class.
-- Refactor decistion tree classes to avoid unneeded type conversion.
+- Refactor decision tree classes to avoid unneeded type conversion.
 
 # 0.6.0
 - Add class for Principal Component Analysis.

data/lib/svmkit/linear_model/sgd_linear_estimator.rb

@@ -79,7 +79,7 @@ module SVMKit
 
       def split_weight(weight)
         if @params[:fit_bias]
-          [weight[0...-1], weight[-1]]
+          [weight[0...-1].dup, weight[-1]]
         else
           [weight, 0.0]
         end

data/lib/svmkit/polynomial_model/factorization_machine_classifier.rb

@@ -253,7 +253,7 @@ module SVMKit
       end
 
       def split_weight_vec_bias(weight_vec)
-        weights = weight_vec[0...-1]
+        weights = weight_vec[0...-1].dup
         bias = weight_vec[-1]
         [weights, bias]
       end

data/lib/svmkit/polynomial_model/factorization_machine_regressor.rb

@@ -185,7 +185,7 @@ module SVMKit
       end
 
       def split_weight_vec_bias(weight_vec)
-        weights = weight_vec[0...-1]
+        weights = weight_vec[0...-1].dup
         bias = weight_vec[-1]
         [weights, bias]
       end

data/lib/svmkit/tree/decision_tree_classifier.rb

@@ -91,8 +91,9 @@ module SVMKit
         n_samples, n_features = x.shape
         @params[:max_features] = n_features if @params[:max_features].nil?
         @params[:max_features] = [@params[:max_features], n_features].min
-        @classes = Numo::Int32.asarray(y.to_a.uniq.sort)
-        build_tree(x, y)
+        uniq_y = y.to_a.uniq.sort
+        @classes = Numo::Int32.asarray(uniq_y)
+        build_tree(x, y.map { |v| uniq_y.index(v) })
         eval_importance(n_samples, n_features)
         self
       end
@@ -174,36 +175,35 @@ module SVMKit
       def build_tree(x, y)
         @n_leaves = 0
         @leaf_labels = []
-        @tree = grow_node(0, x, y)
+        @tree = grow_node(0, x, y, impurity(y))
         @leaf_labels = Numo::Int32[*@leaf_labels]
         nil
       end
 
-      def grow_node(depth, x, y)
-        if @params[:max_leaf_nodes].is_a?(Integer)
+      def grow_node(depth, x, y, whole_impurity)
+        unless @params[:max_leaf_nodes].nil?
           return nil if @n_leaves >= @params[:max_leaf_nodes]
         end
 
         n_samples, n_features = x.shape
-        if @params[:min_samples_leaf].is_a?(Integer)
-          return nil if n_samples <= @params[:min_samples_leaf]
-        end
+        return nil if n_samples <= @params[:min_samples_leaf]
 
-        node = Node.new(depth: depth, impurity: impurity(y), n_samples: n_samples)
+        node = Node.new(depth: depth, impurity: whole_impurity, n_samples: n_samples)
 
         return put_leaf(node, y) if y.to_a.uniq.size == 1
 
-        if @params[:max_depth].is_a?(Integer)
+        unless @params[:max_depth].nil?
           return put_leaf(node, y) if depth == @params[:max_depth]
         end
 
-        feature_id, threshold, left_ids, right_ids, max_gain =
-          rand_ids(n_features).map { |f_id| [f_id, *best_split(x[true, f_id], y)] }.max_by(&:last)
-        return put_leaf(node, y) if max_gain.nil?
-        return put_leaf(node, y) if max_gain.zero?
+        feature_id, threshold, left_ids, right_ids, left_impurity, right_impurity, gain =
+          rand_ids(n_features).map { |f_id| [f_id, *best_split(x[true, f_id], y, whole_impurity)] }.max_by(&:last)
+
+        return put_leaf(node, y) if gain.nil? || gain.zero?
+
+        node.left = grow_node(depth + 1, x[left_ids, true], y[left_ids], left_impurity)
+        node.right = grow_node(depth + 1, x[right_ids, true], y[right_ids], right_impurity)
 
-        node.left = grow_node(depth + 1, x[left_ids, true], y[left_ids])
-        node.right = grow_node(depth + 1, x[right_ids, true], y[right_ids])
         return put_leaf(node, y) if node.left.nil? && node.right.nil?
 
         node.feature_id = feature_id
@@ -213,7 +213,7 @@ module SVMKit
       end
 
       def put_leaf(node, y)
-        node.probs = Numo::DFloat.cast(@classes.map { |c| y.eq(c).count_true }) / node.n_samples
+        node.probs = y.bincount(minlength: @classes.size) / node.n_samples.to_f
         node.leaf = true
         node.leaf_id = @n_leaves
         @n_leaves += 1
@@ -225,27 +225,23 @@ module SVMKit
         [*0...n].sample(@params[:max_features], random: @rng)
       end
 
-      def best_split(features, labels)
+      def best_split(features, labels, whole_impurity)
+        n_samples = labels.size
         features.to_a.uniq.sort.each_cons(2).map do |l, r|
           threshold = 0.5 * (l + r)
-          left_ids, right_ids = splited_ids(features, threshold)
-          [threshold, left_ids, right_ids, gain(labels, labels[left_ids], labels[right_ids])]
+          left_ids = features.le(threshold).where
+          right_ids = features.gt(threshold).where
+          left_impurity = impurity(labels[left_ids])
+          right_impurity = impurity(labels[right_ids])
+          gain = whole_impurity -
+                 left_impurity * left_ids.size.fdiv(n_samples) -
+                 right_impurity * right_ids.size.fdiv(n_samples)
+          [threshold, left_ids, right_ids, left_impurity, right_impurity, gain]
         end.max_by(&:last)
       end
 
-      def splited_ids(features, threshold)
-        [features.le(threshold).where, features.gt(threshold).where]
-      end
-
-      def gain(labels, labels_left, labels_right)
-        prob_left = labels_left.size.fdiv(labels.size)
-        prob_right = labels_right.size.fdiv(labels.size)
-        impurity(labels) - prob_left * impurity(labels_left) - prob_right * impurity(labels_right)
-      end
-
       def impurity(labels)
-        cls = labels.to_a.uniq.sort
-        cls.size == 1 ? 0.0 : send(@criterion, Numo::DFloat[*(cls.map { |c| labels.eq(c).count_true.fdiv(labels.size) })])
+        send(@criterion, labels.bincount / labels.size.to_f)
       end
 
       def gini(posterior_probs)
@@ -253,7 +249,7 @@ module SVMKit
       end
 
       def entropy(posterior_probs)
-        -(posterior_probs * Numo::NMath.log(posterior_probs)).sum
+        -(posterior_probs * Numo::NMath.log(posterior_probs + 1)).sum
       end
 
       def eval_importance(n_samples, n_features)
@@ -269,7 +265,8 @@ module SVMKit
         return nil if node.leaf
         return nil if node.left.nil? || node.right.nil?
         gain = node.n_samples * node.impurity -
-               node.left.n_samples * node.left.impurity - node.right.n_samples * node.right.impurity
+               node.left.n_samples * node.left.impurity -
+               node.right.n_samples * node.right.impurity
         @feature_importances[node.feature_id] += gain
         eval_importance_at_node(node.left)
         eval_importance_at_node(node.right)

data/lib/svmkit/tree/decision_tree_regressor.rb

@@ -151,12 +151,12 @@ module SVMKit
       def build_tree(x, y)
         @n_leaves = 0
         @leaf_values = []
-        @tree = grow_node(0, x, y)
+        @tree = grow_node(0, x, y, impurity(y))
         @leaf_values = Numo::DFloat.cast(@leaf_values)
         nil
       end
 
-      def grow_node(depth, x, y)
+      def grow_node(depth, x, y, whole_impurity)
         unless @params[:max_leaf_nodes].nil?
           return nil if @n_leaves >= @params[:max_leaf_nodes]
         end
@@ -164,7 +164,7 @@ module SVMKit
         n_samples, n_features = x.shape
         return nil if n_samples <= @params[:min_samples_leaf]
 
-        node = Node.new(depth: depth, impurity: impurity(y), n_samples: n_samples)
+        node = Node.new(depth: depth, impurity: whole_impurity, n_samples: n_samples)
 
         return put_leaf(node, y) if (y - y.mean(0)).sum.abs.zero?
 
@@ -172,12 +172,14 @@ module SVMKit
           return put_leaf(node, y) if depth == @params[:max_depth]
         end
 
-        feature_id, threshold, left_ids, right_ids, max_gain =
-          rand_ids(n_features).map { |f_id| [f_id, *best_split(x[true, f_id], y)] }.max_by(&:last)
-        return put_leaf(node, y) if max_gain.nil? || max_gain.zero?
+        feature_id, threshold, left_ids, right_ids, left_impurity, right_impurity, gain =
+          rand_ids(n_features).map { |f_id| [f_id, *best_split(x[true, f_id], y, whole_impurity)] }.max_by(&:last)
+
+        return put_leaf(node, y) if gain.nil? || gain.zero?
+
+        node.left = grow_node(depth + 1, x[left_ids, true], y[left_ids, true], left_impurity)
+        node.right = grow_node(depth + 1, x[right_ids, true], y[right_ids, true], right_impurity)
 
-        node.left = grow_node(depth + 1, x[left_ids, true], y[left_ids, true])
-        node.right = grow_node(depth + 1, x[right_ids, true], y[right_ids, true])
         return put_leaf(node, y) if node.left.nil? && node.right.nil?
 
         node.feature_id = feature_id
@@ -199,24 +201,21 @@ module SVMKit
         [*0...n].sample(@params[:max_features], random: @rng)
       end
 
-      def best_split(features, values)
+      def best_split(features, values, whole_impurity)
+        n_samples = values.shape[0]
         features.to_a.uniq.sort.each_cons(2).map do |l, r|
           threshold = 0.5 * (l + r)
-          left_ids, right_ids = splited_ids(features, threshold)
-          [threshold, left_ids, right_ids, gain(values, values[left_ids], values[right_ids])]
+          left_ids = features.le(threshold).where
+          right_ids = features.gt(threshold).where
+          left_impurity = impurity(values[left_ids, true])
+          right_impurity = impurity(values[right_ids, true])
+          gain = whole_impurity -
+                 left_impurity * left_ids.size.fdiv(n_samples) -
+                 right_impurity * right_ids.size.fdiv(n_samples)
+          [threshold, left_ids, right_ids, left_impurity, right_impurity, gain]
         end.max_by(&:last)
       end
 
-      def splited_ids(features, threshold)
-        [features.le(threshold).where, features.gt(threshold).where]
-      end
-
-      def gain(values, values_left, values_right)
-        prob_left = values_left.shape[0].fdiv(values.shape[0])
-        prob_right = values_right.shape[0].fdiv(values.shape[0])
-        impurity(values) - prob_left * impurity(values_left) - prob_right * impurity(values_right)
-      end
-
       def impurity(values)
         send(@criterion, values)
       end

data/lib/svmkit/version.rb

@@ -3,5 +3,5 @@
 # SVMKit is a machine learning library in Ruby.
 module SVMKit
   # @!visibility private
-  VERSION = '0.6.1'.freeze
+  VERSION = '0.6.2'.freeze
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: svmkit
 version: !ruby/object:Gem::Version
-  version: 0.6.1
+  version: 0.6.2
 platform: ruby
 authors:
 - yoshoku
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2018-09-10 00:00:00.000000000 Z
+date: 2018-09-17 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: numo-narray