hybridforest 0.10.0 → 0.14.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: fd85ad9a8081b634556371599f61557c0490c0694bdc24f8aa8d866d44ba0843
-  data.tar.gz: f58382ac2737bce52c85cdf83e27f3099190a0f52be8c159780033194c183dd8
+  metadata.gz: 4311342fccd332cd6f98b2b6a30f32acc01f2f3090490e452e9bfe8981730f07
+  data.tar.gz: 2a4461a04ac9232d5506271ddbf8b9f9b49b7397a06c00aea894711c3ce0586f
 SHA512:
-  metadata.gz: c6678ba9cf3ccf15970801879a4d66f6ef98c8cf80353fc4ce2e9808f79063c447d19e96be374223403fedc093ac67cdd6599140df391ca68d05c708725ba4fc
-  data.tar.gz: 5c94ea5df0e6305719bf8ae9848a0a36d95d71fa51d7abe7bb09ed3f32720df98616bd3f2892e5c3467b647c9f6ebb7b89486e5cff771dde3141388d026fefe4
+  metadata.gz: 9708316bfa685c814afbffea2d2ea238dd8736ac929bd782a365d6ab65dbe2d1987df7a66750db3d9108071279a99cb1206f179163f1383494c0ebcda704fab6
+  data.tar.gz: efbd6db6210d02830734bb7e3411f8b9553ed8b6172df59f1180e801884057d34edd85a67cd6c63e6b09c8531cf434ccdf265e14f4d8d26e035b333a5805c48d

data/CHANGELOG.md

@@ -34,4 +34,23 @@
 
 ## [0.10.0] - 2021-12-29
 
-- Refactor dataframe extensions
+- Refactor dataframe extensions
+
+## [0.11.0] - 2021-12-29
+
+- Randomize Utils.train_test_split
+- Refactor Utils module
+
+## [0.12.0] - 2022-01-08
+
+- Allow Utils.random_sample to be passed a dataframe or a dataframe convertible object
+- Allow Utils.random_sample's 'size' arg to equal the size of the initial dataframe if the strategy is sampling with replacement
+
+## [0.13.0] - 2022-01-09
+
+- Refactor forest growers
+
+
+## [0.14.0] - 2022-01-09
+
+- Refactor hybrid forest grower

data/Gemfile.lock

@@ -1,7 +1,7 @@
 PATH
   remote: .
   specs:
-    hybridforest (0.9.0)
+    hybridforest (0.13.0)
       activesupport (~> 6.1)
       rake (~> 13.0)
       require_all

data/lib/hybridforest/forests/forest_growers/cart_grower.rb

@@ -8,7 +8,7 @@ module HybridForest
         def grow_forest(instances, number_of_trees)
           forest = []
           number_of_trees.times do
-            sample, _, _ = HybridForest::Utils.train_test_bootstrap_split(instances)
+            sample = HybridForest::Utils.random_sample(data: instances, size: instances.size)
             forest << HybridForest::Trees::CARTTree.new.fit(sample)
           end
           forest

data/lib/hybridforest/forests/forest_growers/hybrid_grower.rb

@@ -11,8 +11,9 @@ module HybridForest
         def grow_forest(instances, number_of_trees)
           forest = []
           number_of_trees.times do
-            in_of_bag, out_of_bag, out_of_bag_labels = HybridForest::Utils.train_test_bootstrap_split(instances)
-            tree_results = grow_trees(TREE_TYPES, in_of_bag, out_of_bag, out_of_bag_labels)
+            iob_data, oob_data, oob_labels = HybridForest::Utils.train_test_bootstrap_split(instances)
+            trees = grow_trees(TREE_TYPES, iob_data)
+            tree_results = predict_evaluate_trees(trees, oob_data, oob_labels)
             best_tree = select_best_tree(tree_results)
             forest << best_tree
           end
@@ -21,25 +22,28 @@ module HybridForest
 
         private
 
-        def fit_and_predict(tree_class, in_of_bag, out_of_bag, out_of_bag_labels)
-          tree = tree_class.new.fit(in_of_bag)
-          tree_predictions = tree.predict(out_of_bag)
-          tree_accuracy = HybridForest::Utils.accuracy(tree_predictions, out_of_bag_labels)
-          {tree: tree, oob_accuracy: tree_accuracy}
+        def grow_trees(tree_types, iob_data)
+          tree_types.collect do |tree_type|
+            tree_type.new.fit(iob_data)
+          end
+        end
+
+        def predict_evaluate_trees(trees, oob_data, oob_labels)
+          trees.collect do |tree|
+            predict_evaluate(tree, oob_data, oob_labels)
+          end
+        end
+
+        def predict_evaluate(tree, data, actual_labels)
+          predicted_labels = tree.predict(data)
+          accuracy = HybridForest::Utils.accuracy(predicted_labels, actual_labels)
+          {tree: tree, oob_accuracy: accuracy}
         end
 
         def select_best_tree(tree_results)
           best_result = tree_results.max_by(1) { |result| result[:oob_accuracy] }.first
           best_result[:tree]
         end
-
-        def grow_trees(tree_types, in_of_bag, out_of_bag, out_of_bag_labels)
-          tree_results = []
-          tree_types.each do |tree_type|
-            tree_results << fit_and_predict(tree_type, in_of_bag, out_of_bag, out_of_bag_labels)
-          end
-          tree_results
-        end
       end
     end
   end

data/lib/hybridforest/forests/forest_growers/id3_grower.rb

@@ -8,7 +8,7 @@ module HybridForest
         def grow_forest(instances, number_of_trees)
           forest = []
           number_of_trees.times do
-            sample, _, _ = HybridForest::Utils.train_test_bootstrap_split(instances)
+            sample = HybridForest::Utils.random_sample(data: instances, size: instances.size)
             forest << HybridForest::Trees::ID3Tree.new.fit(sample)
           end
           forest

data/lib/hybridforest/trees/feature_selectors/random_feature_subspace.rb

@@ -6,6 +6,7 @@ module HybridForest
   module Trees
     class RandomFeatureSubspace
       def select_features(all_features)
+        # TODO: Allow the subspace size to be configured.
         n = default_subspace_size(all_features.count)
         indices = Set.new
         until indices.size == n

data/lib/hybridforest/utilities/utils.rb

@@ -12,17 +12,16 @@ module HybridForest
     # of independent features and an array of labels. Returns [+training_set+, +testing_set+, +testing_set_labels+]
     #
     def self.train_test_split(dataset, test_set_size = 0.20)
-      # TODO: Shuffle and stratify samples
+      # TODO: Offer stratify param
       dataset = to_dataframe(dataset)
+      all_rows = (0...dataset.count).to_a
 
       test_set_count = (dataset.count * test_set_size).floor
-      test_set_indices = 0..test_set_count
-      test_set = dataset[test_set_indices]
-      test_set_labels = test_set.class_labels
-      test_set.except!(test_set.label)
+      test_set_rows = rand_uniq_nums(test_set_count, 0...dataset.count)
+      test_set = dataset[test_set_rows]
+      test_set, test_set_labels = test_set.disconnect_labels
 
-      train_set_indices = test_set_count + 1...dataset.count
-      train_set = dataset[train_set_indices]
+      train_set = dataset[all_rows - test_set_rows]
 
       [train_set, test_set, test_set_labels]
     end
@@ -37,20 +36,13 @@ module HybridForest
       dataset = to_dataframe(dataset)
       all_rows = (0...dataset.count).to_a
 
-      train_set = Rover::DataFrame.new
-      train_set_rows = []
-      dataset.count.times do
-        row = all_rows.sample
-        train_set_rows << row
-        train_set.concat(dataset[row])
-      end
+      train_set_rows = rand_nums(dataset.count, 0...dataset.count)
+      train_set = dataset[train_set_rows]
 
       return train_test_split(dataset) if train_set_rows.sort == all_rows
 
-      test_set_rows = all_rows - train_set_rows
-      test_set = dataset[test_set_rows]
-      test_set_labels = test_set.class_labels
-      test_set.except!(test_set.label)
+      test_set = dataset[all_rows - train_set_rows]
+      test_set, test_set_labels = test_set.disconnect_labels
 
       [train_set, test_set, test_set_labels]
     end
@@ -86,18 +78,18 @@ module HybridForest
     # Draws a random sample of +size+ from +data+.
     #
     def self.random_sample(data:, size:, with_replacement: true)
-      raise ArgumentError, "Invalid sample size" if size < 1 || size > data.count
+      data = to_dataframe(data)
 
-      if with_replacement
-        rows = size.times.collect { rand(0...data.count) }
-        data[rows]
+      if size < 1 || (!with_replacement && size > data.count)
+        raise ArgumentError, "Invalid sample size"
+      end
+
+      rows = if with_replacement
+        rand_nums(size, 0...data.count)
       else
-        rows = Set.new
-        until rows.size == size
-          rows << rand(0...data.count)
-        end
-        data[rows.to_a]
+        rand_uniq_nums(size, 0...data.count)
       end
+      data[rows]
     end
 
     # Outputs a report of common prediction metrics.
@@ -168,6 +160,12 @@ module HybridForest
         def class_labels
           self[label].to_a
         end
+
+        def disconnect_labels
+          labels = class_labels
+          except!(label)
+          [self, labels]
+        end
       end
     end
 
@@ -202,5 +200,23 @@ module HybridForest
     def false_label?(label)
       [false, 0].include? label
     end
+
+    ##
+    # Returns an array of +n+ random numbers in the exclusive +range+.
+    def rand_nums(n, range)
+      n.times.collect { rand(range) }
+    end
+
+    ##
+    # Returns an array of +n+ _unique_ random numbers in the exclusive +range+.
+    def rand_uniq_nums(n, range)
+      raise ArgumentError if n > range.size
+
+      nums = Set.new
+      until nums.size == n
+        nums << rand(range)
+      end
+      nums.to_a
+    end
   end
 end

data/lib/hybridforest/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module HybridForest
-  VERSION = "0.10.0"
+  VERSION = "0.14.0"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: hybridforest
 version: !ruby/object:Gem::Version
-  version: 0.10.0
+  version: 0.14.0
 platform: ruby
 authors:
 - hi-tech-jazz
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2021-12-29 00:00:00.000000000 Z
+date: 2022-01-09 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: rake